fix(autocompact): force-compaction through breaker + hard message-count cap (closes #1373)

[midlaj-muhammed]

Fixes #1373.

What changed

• src/services/compact/autoCompact.ts — autoCompactIfNeeded now honors forceReason even when the breaker is in cool-down. The breaker still increments consecutiveFailures and re-trips on a forced failure, so retry-storm protection is preserved.
• src/services/compact/autoCompact.ts — new MAX_ACTIVE_MESSAGES_HARD_CAP = 1000 constant + getMaxActiveMessagesHardCap() helper. Override via OPENCLAUDE_MAX_ACTIVE_MESSAGES_HARD_CAP; 0 disables.
• src/query.ts — the existing opt-in maxMessagesCompactionThreshold check now also consults the new hard cap. Uses the tighter of the two.
• src/services/compact/compactWarningState.ts — new BreakerTripState store + recordBreakerTripped / getBreakerTripState / clearBreakerTrippedState exports (REPL/SDK wiring deferred to a follow-up PR).
• src/services/compact/autoCompact.ts — calls recordBreakerTripped() when the breaker trips.

Why

Issue #1373 reports an OOM after ~3h of normal use. Root cause: a single summarization failure trips the breaker, which then blocks the only path that would shrink state.messages while every new turn keeps appending. The user's stated expected behavior was:

1. Auto-compact should re-attempt after a cooldown period instead of being permanently disabled.
2. An absolute cap on state.messages should prevent runaway memory growth.

(1) is partly already implemented (the half-open breaker does re-attempt); the missing piece is honoring forced compaction through the breaker. (2) is the new hard cap.

Impact

• Long-running sessions with a wedged summarization now have a guaranteed compaction trigger at 1000 messages.
• Existing maxMessagesCompactionThreshold users see no behavior change for their setting.
• Provider-agnostic — Anthropic, OpenAI-compat, Gemini, GitHub Models, Codex, Ollama, etc. all benefit.

Checks

• bun run typecheck — clean
• bun run typecheck:type-tests — 10 files pass
• bun run security:pr-scan — clean
• bun run verify:privacy — all banned-pattern checks pass
• bun run smoke — builds, 0.18.0 (OpenClaude)
• bun test src/services/compact/ — 72 pass (was 64, +8 new)
• bun test src/query/autoCompactCooldown.test.ts — 6 pass (was 5, +1 new)
• bun test src/QueryEngine.autoCompactCooldown.test.ts — 1 pass
• bun run test:full — 14 pre-existing failures (verified identical with changes stashed); all are unrelated teammate-routing + /export env-dependent tests on main.

Follow-up

• Wire getBreakerTripState() into the REPL/SDK so users see "auto-compact paused" instead of silent failure. The store is in place.
• Optional per-model hard cap.

🤖 Generated with Claude Code

Summary by CodeRabbit

• New Features
  • Added an operator hard cap for active message counts (OPENCLAUDE_MAX_ACTIVE_MESSAGES_HARD_CAP, default 1000; 0 disables) that can trigger auto-compaction independently of user settings.
  • Added session-scoped auto-compaction circuit-breaker tracking, with automatic recovery after successful compactions and manual /compact.
• Bug Fixes
  • Improved forced auto-compaction cooldown handling to prevent repeated forced retriggers; forced attempts are managed to avoid cooldown “storm” behavior.
  • In-swarm runs, compaction boundary messages now reset buffered context before continuing.
• Documentation
  • Updated advanced setup and Config UI copy to clearly separate user-threshold vs operator hard-cap behavior.
• Tests
  • Added regression coverage for Issue #1373, including hard-cap/cooldown interactions and session isolation/reset behavior.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Pro Plus

Run ID: e457550b-968b-4316-bc0c-55d4394bb21a

📥 Commits

Reviewing files that changed from the base of the PR and between caf86b4 and 553d92e.

📒 Files selected for processing (1)

• src/query.ts

📜 Recent review details

🧰 Additional context used

📓 Path-based instructions (6)

src/**/*.{ts,tsx}

📄 CodeRabbit inference engine (AGENTS.md)

Use TypeScript with strict mode and ESM imports

Files:

• src/query.ts

**/*.{ts,tsx,js,jsx,py,json,md}

📄 CodeRabbit inference engine (CONTRIBUTING.md)

Follow the existing code style in the touched files

Files:

• src/query.ts

**/*.{ts,tsx,js,jsx,py}

📄 CodeRabbit inference engine (CONTRIBUTING.md)

Keep comments useful and concise

Files:

• src/query.ts

**/*.{ts,tsx}

📄 CodeRabbit inference engine (CONTRIBUTING.md)

Follow TypeScript strict mode and type safety practices by running typecheck before submitting

Files:

• src/query.ts

**/*

⚙️ CodeRabbit configuration file

**/*: Apply the OpenClaude maintainer review rubric from AGENTS.md. Review the current diff, not stale discussion context. Separate real blockers from suggestions. Do not request changes for vague style churn. Treat approval as merge-ready from CodeRabbit's side, pending required human review and GitHub Checks. If checks are failing or unavailable, say so clearly instead of implying the PR is fully ready.

Files:

• src/query.ts

**

⚙️ CodeRabbit configuration file

**: # AGENTS.md - AI Agent Coding Guide

This guide is for AI coding agents working in the OpenClaude repository. Read it before changing code, and also follow CONTRIBUTING.md for contributor policy, PR expectations, review follow-up, and project scope.

Project Snapshot

OpenClaude is a coding-agent CLI for cloud and local model providers. It supports OpenAI-compatible APIs, Anthropic, Gemini, DeepSeek, Ollama, MCP, local backends, slash commands, tools, agents, and a React/Ink terminal UI.

The installed CLI runs on Node.js >=22.0.0. Bun is used for source builds, scripts, dependency management, and tests.

Work Style

• Keep changes focused on one problem.
• Prefer existing patterns in the file or nearby module.
• Avoid unrelated formatting, renames, dependency changes, or broad rewrites.
• Add or update tests when behavior changes.
• Update docs when setup, commands, provider behavior, or user-facing behavior changes.
• For new features, larger refactors, dependencies, or runtime changes, follow the issue-first guidance in CONTRIBUTING.md.

Stack And Conventions

• TypeScript with strict mode and ESM imports.
• React + Ink for terminal UI.
• Bun lockfile and Bun scripts for development workflows.
• Node runtime for the built CLI.
• Python exists for legacy/local-provider helper code. Do not add new Python code or expand Python-based features unless a maintainer explicitly approves that direction.

Common libraries and patterns:

• chalk for terminal color.
• commander for CLI argument parsing.
• execa for child processes.
• Existing service, provider, settings, permission, and UI patterns over new abstractions.

Repository Map

• src/commands/ - slash and CLI command implementations.
• src/components/ - React/Ink UI components.
• src/services/ - API, MCP, OAuth, wiki, voice, and other service integrations.
• src/tools/ - tool implementations.
• src/utils/ - shared utilities.
• `src/integration...

Files:

• src/query.ts

🔇 Additional comments (1)

src/query.ts (1)

983-1000: LGTM!

---

📝 Walkthrough

Walkthrough

Adds a hard operator cap for active messages (default 1000, env override OPENCLAUDE_MAX_ACTIVE_MESSAGES_HARD_CAP), persists auto-compact breaker-trip state independently from warning suppression, refactors autoCompactIfNeeded to allow forced compactions to bypass circuit-breaker cooldown while preserving failure bookkeeping, enforces the tighter nonzero cap from the query loop with cooldown gating to prevent retry storms after forced-attempt failures, and syncs in-process teammate message history with compaction boundaries.

Changes

Hard Message Cap and Circuit Breaker Override

Layer / File(s)	Summary
Breaker trip state store and cleanup integration src/services/compact/compactWarningState.ts, src/services/compact/postCompactCleanup.ts, src/commands/compact/compact.ts, src/services/compact/breakerTripReset.test.ts	New BreakerTripState type and independent breakerTripStore session-scoped storage separate from compactWarningStore; recordBreakerTripped idempotently marks breaker tripped while preserving original trippedAtMs on retries; getBreakerTripState and clearBreakerTrippedState accessors for reading/resetting. Compact command imports and calls clearBreakerTrippedState after successful manual /compact runs (session-memory, traditional, reactive paths). Post-compaction cleanup JSDoc clarifies breaker-trip store is not implicitly cleared. Codified test contract: runPostCompactCleanup() does not clear breaker-trip state (cache-only and subagent cleanup); clearBreakerTrippedState() correctly transitions breaker and preserves store identity; session isolation prevents subagent trips from affecting main thread.
Hard message cap constant and resolver src/services/compact/autoCompact.ts, src/services/compact/autoCompact.test.ts	MAX_ACTIVE_MESSAGES_HARD_CAP constant (default 1000) with env override OPENCLAUDE_MAX_ACTIVE_MESSAGES_HARD_CAP (strict non-negative, 0 disables); getMaxActiveMessagesHardCap() resolver with fallback for invalid env. Test helper importAutoCompact extended to accept autoCompactEnabled override. Tests cover default constant, positive override, disable via 0, and fallback for non-numeric/negative values.
shouldAutoCompact forceReason parameter src/services/compact/autoCompact.ts	AutoCompactTrackingState adds lastForcedFailureAtMs?: number to distinguish forced failures. Update shouldAutoCompact signature to accept `forceReason?: 'memory-pressure'
autoCompactIfNeeded forced-compaction bypass and breaker bookkeeping src/services/compact/autoCompact.ts, src/services/compact/autoCompact.test.ts	autoCompactIfNeeded refactored so forceReason bypasses cooldown early-skip (breakerState.action === 'skip' && !isForced), enabling forced attempts despite cooldown. Derive wasHalfOpen and effectiveConsecutiveFailures for correct failure bookkeeping; restrict half-open reset to half-open, non-forced attempts. Use derived values for increment/user-abort logic. Call recordBreakerTripped when breaker trips, guarded by isMainThreadCompact(querySource). Return lastForcedFailureAtMs only for forced attempts (otherwise omitted). Tests validate one-shot forceReason consumption, circuit breaker re-trip with metadata bounds, memory-pressure forced bypass, non-forced failures leaving lastForcedFailureAtMs undefined, and forced compaction (hard-message-count, memory-pressure) bypassing DISABLE_AUTO_COMPACT/autoCompactEnabled: false/DISABLE_COMPACT while user-message-count respects guards. Subagent isolation tests ensure subagent compactions don't corrupt main-thread breaker-trip store.
Query loop two-tier cap enforcement and cooldown gating src/query.ts, src/query/autoCompactCooldown.test.ts, src/utils/memoryPressure.ts, docs/advanced-setup.md, src/components/Settings/Config.tsx	Query imports hard-cap resolver and cooldown timer. Computes userCap from config/OPENCLAUDE_MAX_ACTIVE_MESSAGES (or disable); fetches hardCap; derives activeCap as tighter nonzero value (0 disabled). Gates re-trigger via forcedCooldownActive from lastForcedFailureAtMs + cooldown window, honoring nextRetryAtMs only when consistent. Forces compaction with forceReason: 'hard-message-count' or 'user-message-count' when message length exceeds activeCap and cooldown inactive; logs distinguish cap vs cooldown deferral. Gates memory-pressure consumeCompactionRequest() re-arm to avoid re-triggering during cooldown. Threads lastForcedFailureAtMs through autocompact result destructuring and nextTracking propagation, preserving prior value unless autocompact returns fresh. Memory-pressure docs clarify cooldown gating. Extended SAVED_ENV for cap/cooldown env vars with explicit beforeEach cleanup. Issue #1373 regressions: hard-cap forcing overrides breaker, operator-cap-disabled (user cap when hardCap disabled), cooldown-suppresses-retry (forced failure blocks re-trigger), cooldown-elapses (cap re-trigger after elapsed cooldown), lastForcedFailureAtMs-persists (prevents retry storms), DISABLE_AUTO_COMPACT/DISABLE_COMPACT bypass (forced compaction runs), cooldown-gate-stability (hard-cap cooldown gate holds when non-forced skip clears nextRetryAtMs), memory-pressure cooldown-gate suppression, and memory-pressure forceReason precedence (overrides user-message-count when hardCap disabled; preserves hard-message-count priority when enabled). Updated documentation and Config dialog to describe the two-tier model and clarify hard-cap independence from user opt-outs.
In-process teammate message sync on compaction boundary src/utils/swarm/inProcessRunner.ts	In runInProcessTeammate, message loop detects system/compact_boundary emitted by runAgent and clears the allMessages buffer before appending the boundary message. Keeps teammate message history aligned with compaction state, preventing unbounded growth of per-teammate accumulator.

Estimated code review effort

🎯 4 (Complex) | ⏱️ ~55 minutes

Possibly related issues

• Gitlawb/openclaude#1379: Implements the hard message-count cap and in-process teammate message retention strategy identified in that follow-up, including getMaxActiveMessagesHardCap() constant and compact_boundary detection in inProcessRunner.ts.

Possibly related PRs

• Gitlawb/openclaude#1587: Both modify message-count compaction forcing logic in src/query.ts around maxMessagesCompactionThreshold and OPENCLAUDE_MAX_ACTIVE_MESSAGES threshold handling.

Suggested labels

bug

Suggested reviewers

• jatmn
• kevincodex1

🚥 Pre-merge checks | ✅ 5 | ❌ 2

❌ Failed checks (2 warnings)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 47.37% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.
Risk Surface Disclosed	⚠️ Warning	PR introduces new env var OPENCLAUDE_MAX_ACTIVE_MESSAGES_HARD_CAP affecting startup/config behavior, but provided materials lack explicit risk surface review call-out. Outstanding review comments...	Verify that review comments from @jatmn address the risk surface of the new environment variable (startup/config impact, default 1000 message cap, disable via =0).

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	Title clearly describes the main change: forced-compaction through the breaker and new hard message-count cap, directly addressing issue #1373.
Description check	✅ Passed	Description covers all required sections: what changed, why it changed, impact, and testing. Includes checks performed and follow-up work.
Linked Issues check	✅ Passed	Changes fully address #1373's two stated requirements: (1) forced compaction re-attempts through breaker cooldown via forceReason honor, and (2) hard cap (1000 messages) prevents unbounded growth.
Out of Scope Changes check	✅ Passed	All changes are tightly scoped to issue #1373: breaker-aware forcing, hard cap implementation, cooldown gating, and supporting infrastructure. Documentation and test additions are justified by the core fix.
No Hidden Policy Change	✅ Passed	All changes are scoped to message compaction triggering and circuit-breaker state tracking. No hidden policy, telemetry, routing, auth, or trust-model changes detected.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[coderabbitai]

Actionable comments posted: 1

Caution

Some comments are outside the diff and can’t be posted inline due to platform limitations.

⚠️ Outside diff range comments (1)

src/query.ts (1)

572-606: ⚠️ Potential issue | 🔴 Critical | ⚡ Quick win

activeCap logic breaks when user cap is set but hard cap is disabled.

The current logic on line 590:

const activeCap = userCap > 0 ? Math.min(userCap, hardCap) : hardCap

fails when userCap=100 and hardCap=0 (operator disables safety net via OPENCLAUDE_MAX_ACTIVE_MESSAGES_HARD_CAP=0, but user sets opt-in threshold). Current result: activeCap = Math.min(100, 0) = 0, which disables forced compaction and ignores the user's explicit preference.

The user cap and hard cap should be independent: disabling the operator-level safety net shouldn't break user-level compaction preferences.

🔧 Proposed fix

-      const activeCap = userCap > 0 ? Math.min(userCap, hardCap) : hardCap
+      let activeCap = 0
+      if (userCap > 0 && hardCap > 0) {
+        activeCap = Math.min(userCap, hardCap)
+      } else if (userCap > 0) {
+        activeCap = userCap
+      } else {
+        activeCap = hardCap
+      }

Or more concisely:

-      const activeCap = userCap > 0 ? Math.min(userCap, hardCap) : hardCap
+      const activeCap =
+        userCap > 0 && hardCap > 0
+          ? Math.min(userCap, hardCap)
+          : userCap > 0
+            ? userCap
+            : hardCap

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/query.ts` around lines 572 - 606, The activeCap calculation treats
hardCap=0 as stricter than a positive userCap and thus disables user-configured
compaction; change the logic in the block that computes activeCap (after
getMaxActiveMessagesHardCap(), normalizeMaxMessagesCompactionThreshold(), and
userCap) so that hardCap=0 means "disabled" rather than minimal: implement
something like - if (userCap > 0 && hardCap > 0) activeCap = Math.min(userCap,
hardCap); else activeCap = userCap > 0 ? userCap : hardCap; - this preserves the
min when both are set, uses the userCap when operator hard cap is disabled, and
falls back to hardCap when no userCap is provided.

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@src/query/autoCompactCooldown.test.ts`:
- Around line 442-517: Add a new test mirroring "forced message-count compaction
overrides an active cool-down" that specifically covers the user-cap-only
scenario: set process.env.OPENCLAUDE_MAX_ACTIVE_MESSAGES_HARD_CAP = '0' and
process.env.OPENCLAUDE_MAX_ACTIVE_MESSAGES = '5' (or similar >0), build messages
>5 using overAutoCompactThresholdMessage(), call query(...) with the same
deps/mocks (mock autocompact should assert tracking.forceReason ===
'message-count' and that breaker behavior is correct), use drain(query(...)) and
assert the call model ran (terminal.reason === 'max_turns'), seenTracking
contains the forced message-count reason, and no isApiErrorMessage was yielded;
follow the existing test structure and helper functions (query, drain,
overAutoCompactThresholdMessage, autoCompactTracking shape) to keep assertions
consistent.

---

Outside diff comments:
In `@src/query.ts`:
- Around line 572-606: The activeCap calculation treats hardCap=0 as stricter
than a positive userCap and thus disables user-configured compaction; change the
logic in the block that computes activeCap (after getMaxActiveMessagesHardCap(),
normalizeMaxMessagesCompactionThreshold(), and userCap) so that hardCap=0 means
"disabled" rather than minimal: implement something like - if (userCap > 0 &&
hardCap > 0) activeCap = Math.min(userCap, hardCap); else activeCap = userCap >
0 ? userCap : hardCap; - this preserves the min when both are set, uses the
userCap when operator hard cap is disabled, and falls back to hardCap when no
userCap is provided.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Pro Plus

Run ID: 71cadbd9-2f24-4122-9025-f6070b732ee2

📥 Commits

Reviewing files that changed from the base of the PR and between a3a3c36 and 2f9a3a0.

📒 Files selected for processing (5)

• src/query.ts
• src/query/autoCompactCooldown.test.ts
• src/services/compact/autoCompact.test.ts
• src/services/compact/autoCompact.ts
• src/services/compact/compactWarningState.ts

📜 Review details

🧰 Additional context used

📓 Path-based instructions (5)

**/*.{ts,tsx,js,jsx,py}

📄 CodeRabbit inference engine (CONTRIBUTING.md)

**/*.{ts,tsx,js,jsx,py}: Follow the existing code style in the touched files
Keep comments useful and concise

Files:

• src/services/compact/compactWarningState.ts
• src/services/compact/autoCompact.test.ts
• src/query/autoCompactCooldown.test.ts
• src/services/compact/autoCompact.ts
• src/query.ts

**/*.{ts,tsx}

📄 CodeRabbit inference engine (CONTRIBUTING.md)

Typecheck TypeScript code before submitting (use bun run typecheck)

Files:

• src/services/compact/compactWarningState.ts
• src/services/compact/autoCompact.test.ts
• src/query/autoCompactCooldown.test.ts
• src/services/compact/autoCompact.ts
• src/query.ts

**/*

⚙️ CodeRabbit configuration file

**/*: Apply the OpenClaude maintainer review rubric from AGENTS.md. Review the current diff, not stale discussion context. Separate real blockers from suggestions. Do not request changes for vague style churn. Treat approval as merge-ready from CodeRabbit's side, pending required human review and GitHub Checks. If checks are failing or unavailable, say so clearly instead of implying the PR is fully ready.

Files:

• src/services/compact/compactWarningState.ts
• src/services/compact/autoCompact.test.ts
• src/query/autoCompactCooldown.test.ts
• src/services/compact/autoCompact.ts
• src/query.ts

**

⚙️ CodeRabbit configuration file

**: # Contributing to OpenClaude

Thanks for contributing.

OpenClaude is a fast-moving open-source coding-agent CLI with support for multiple providers, local backends, MCP, and a terminal-first workflow. The best contributions here are focused, well-tested, and easy to review.

Before You Start

• Search existing issues and discussions before opening a new thread.
• Check open pull requests for work that overlaps with your contribution. If a PR already exists that addresses the same change, open an issue or discussion first to align on direction — duplicate PRs may be closed without review.
• Use issues for confirmed bugs and actionable feature work.
• Use discussions for setup help, ideas, and general community conversation.
• For larger changes, open an issue first so the scope is clear before implementation.
• For security reports, follow SECURITY.md.

Pull Requests

Every PR needs a reason. Your PR description must include:

• what changed and why
• the user or developer impact
• the exact checks you ran
• a linked issue when one exists, using Fixes #123, `Closes `#123, or another clear link
• screenshots when the PR touches UI, terminal presentation, or the VS Code extension
• which provider path was tested when the PR changes provider behavior

The PR author is responsible for ensuring their PR is merge-ready. PRs with merge conflicts will not be reviewed or approved until the conflicts are resolved.

Issues are the recommended starting point for anything non-trivial — opening one first helps avoid wasted effort if the change is out of scope or already being worked on. Small fixes, doc corrections, and obvious improvements can stand on their own without a linked issue, as long as the PR description explains the intent.

What Gets Closed Without Review

PRs may be closed without review...

Files:

• src/services/compact/compactWarningState.ts
• src/services/compact/autoCompact.test.ts
• src/query/autoCompactCooldown.test.ts
• src/services/compact/autoCompact.ts
• src/query.ts

{src/**/*.test.ts,src/**/*.test.tsx,tests/**,scripts/**/*.test.ts,vscode-extension/**/*.test.js}

⚙️ CodeRabbit configuration file

{src/**/*.test.ts,src/**/*.test.tsx,tests/**,scripts/**/*.test.ts,vscode-extension/**/*.test.js}: Review tests for meaningful coverage of the changed behavior, isolation of global/env/config state, async cleanup, fake timers, provider profile leaks, and Windows-compatible assumptions. Block when risky runtime changes lack focused regression coverage or tests assert implementation details while missing the user-visible behavior.

Files:

• src/services/compact/autoCompact.test.ts
• src/query/autoCompactCooldown.test.ts

🔇 Additional comments (7)

src/services/compact/compactWarningState.ts (1)

19-73: LGTM!

src/services/compact/autoCompact.ts (3)

90-139: LGTM!

---

436-481: LGTM!

---

603-609: LGTM!

src/services/compact/autoCompact.test.ts (1)

757-928: LGTM!

src/query/autoCompactCooldown.test.ts (1)

27-30: LGTM!

src/query.ts (1)

578-586: Avoid concern: maxMessagesCompactionThreshold is already constrained; invalid values won’t reach parseInt

• normalizeMaxMessagesCompactionThreshold() only allows 'off' | '100' | '200' | '500' | '1000' and maps anything else to 'off', so "abc" in config won’t produce NaN.
• If OPENCLAUDE_MAX_ACTIVE_MESSAGES parses to NaN, userCap > 0 is false and activeCap falls back to hardCap (so no crash; behavior degrades to the safety net cap).

[jatmn]

I found an issue that needs to be addressed before this is ready.

Findings

• [P2] Preserve cooldown after a failed forced hard-cap compaction
  src/query.ts:600
  When a conversation remains over the new hard cap and the forced compaction fails, autoCompactIfNeeded returns a fresh nextRetryAtMs, but the next turn immediately re-enters this block and sets forceReason: 'message-count' again. Because autoCompactIfNeeded skips the breaker cooldown whenever forceReason is present, a persistent summarization/provider failure now retries compaction on every over-cap turn instead of waiting for the recorded cooldown. That contradicts the stated retry-storm protection and can add repeated latency/API attempts in exactly the failure mode this breaker is meant to contain. Please keep the hard-cap recovery path from bypassing an active nextRetryAtMs after a failed forced attempt, or add equivalent backoff before re-forcing.

[kevincodex1]

please address comments by @jatmn

[coderabbitai]

Actionable comments posted: 2

Caution

Some comments are outside the diff and can’t be posted inline due to platform limitations.

⚠️ Outside diff range comments (1)

src/query.ts (1)

631-657: ⚠️ Potential issue | 🟠 Major | ⚡ Quick win

The hard cap still disappears when auto-compact is disabled.

This branch sets forceReason: 'message-count', but shouldAutoCompact(..., true) still returns early on !isAutoCompactEnabled() in src/services/compact/autoCompact.ts Lines 333-375. With DISABLE_AUTO_COMPACT=1 or autoCompactEnabled=false, an over-cap session neither compacts nor blocks, so the new "always active" safety net is effectively off and the message array can keep growing past the hard cap.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/query.ts` around lines 631 - 657, The hard cap safety net is ineffective
when auto-compact is disabled because the shouldAutoCompact function in
src/services/compact/autoCompact.ts returns early when isAutoCompactEnabled() is
false, ignoring the forceReason flag set in the query.ts code. To fix this,
modify the shouldAutoCompact function to check whether forceReason is set
(indicating a forced compaction due to hard cap violation) and bypass the early
return for disabled auto-compact when forceReason is present, ensuring the hard
cap is always enforced as a critical safety mechanism regardless of the
autoCompactEnabled setting.

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@src/query.ts`:
- Around line 782-790: The code in the conditional block handling
lastForcedFailureAtMs is prematurely clearing the timestamp when
autoCompactIfNeeded() doesn't return it. Since autoCompactIfNeeded() only
includes lastForcedFailureAtMs on forced failures and not on subsequent
non-forced skips, the else branch that deletes this field causes the cooldown
timestamp to be lost after the first skipped turn. Instead of unconditionally
deleting lastForcedFailureAtMs when it's undefined, preserve the previous
timestamp from nextTracking unless you have explicit confirmation that the
cooldown window has actually expired (such as successful compaction completion
or verification that the cooldown duration has fully elapsed). This ensures the
forced failure timestamp persists across multiple turns until the cooldown
period genuinely ends.

In `@src/query/autoCompactCooldown.test.ts`:
- Around line 599-602: The test file sets the environment variable
OPENCLAUDE_AUTOCOMPACT_FAILURE_COOLDOWN_MS in the test cases at lines 599-602
and 685-687, but this variable is not being restored after each test, causing a
60-second cooldown to leak into subsequent test suites and cause unrelated tests
to fail. Add OPENCLAUDE_AUTOCOMPACT_FAILURE_COOLDOWN_MS to the SAVED_ENV object
at the top of the file to capture its original value, then ensure the beforeEach
hook clears this variable along with the other environment knobs so that each
test starts with a clean state and prevents test pollution.

---

Outside diff comments:
In `@src/query.ts`:
- Around line 631-657: The hard cap safety net is ineffective when auto-compact
is disabled because the shouldAutoCompact function in
src/services/compact/autoCompact.ts returns early when isAutoCompactEnabled() is
false, ignoring the forceReason flag set in the query.ts code. To fix this,
modify the shouldAutoCompact function to check whether forceReason is set
(indicating a forced compaction due to hard cap violation) and bypass the early
return for disabled auto-compact when forceReason is present, ensuring the hard
cap is always enforced as a critical safety mechanism regardless of the
autoCompactEnabled setting.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Pro Plus

Run ID: ae6275c8-1c7f-4be4-b48c-f25ccf7ac105

📥 Commits

Reviewing files that changed from the base of the PR and between 069d4e6 and a8a2321.

📒 Files selected for processing (4)

• src/query.ts
• src/query/autoCompactCooldown.test.ts
• src/services/compact/autoCompact.test.ts
• src/services/compact/autoCompact.ts

📜 Review details

🧰 Additional context used

📓 Path-based instructions (5)

**/*.{ts,tsx,js,jsx,py}

📄 CodeRabbit inference engine (CONTRIBUTING.md)

**/*.{ts,tsx,js,jsx,py}: Follow the existing code style in the touched files
Keep comments useful and concise

Files:

• src/query/autoCompactCooldown.test.ts
• src/query.ts
• src/services/compact/autoCompact.test.ts
• src/services/compact/autoCompact.ts

**/*.{ts,tsx}

📄 CodeRabbit inference engine (CONTRIBUTING.md)

Typecheck TypeScript code before submitting (use bun run typecheck)

Files:

• src/query/autoCompactCooldown.test.ts
• src/query.ts
• src/services/compact/autoCompact.test.ts
• src/services/compact/autoCompact.ts

**/*

⚙️ CodeRabbit configuration file

**/*: Apply the OpenClaude maintainer review rubric from AGENTS.md. Review the current diff, not stale discussion context. Separate real blockers from suggestions. Do not request changes for vague style churn. Treat approval as merge-ready from CodeRabbit's side, pending required human review and GitHub Checks. If checks are failing or unavailable, say so clearly instead of implying the PR is fully ready.

Files:

• src/query/autoCompactCooldown.test.ts
• src/query.ts
• src/services/compact/autoCompact.test.ts
• src/services/compact/autoCompact.ts

{src/**/*.test.ts,src/**/*.test.tsx,tests/**,scripts/**/*.test.ts,vscode-extension/**/*.test.js}

⚙️ CodeRabbit configuration file

{src/**/*.test.ts,src/**/*.test.tsx,tests/**,scripts/**/*.test.ts,vscode-extension/**/*.test.js}: Review tests for meaningful coverage of the changed behavior, isolation of global/env/config state, async cleanup, fake timers, provider profile leaks, and Windows-compatible assumptions. Block when risky runtime changes lack focused regression coverage or tests assert implementation details while missing the user-visible behavior.

Files:

• src/query/autoCompactCooldown.test.ts
• src/services/compact/autoCompact.test.ts

**

⚙️ CodeRabbit configuration file

**: # Contributing to OpenClaude

Thanks for contributing.

OpenClaude is a fast-moving open-source coding-agent CLI with support for multiple providers, local backends, MCP, and a terminal-first workflow. The best contributions here are focused, well-tested, and easy to review.

Before You Start

• Search existing issues and discussions before opening a new thread.
• Check open pull requests for work that overlaps with your contribution. If a PR already exists that addresses the same change, open an issue or discussion first to align on direction — duplicate PRs may be closed without review.
• Use issues for confirmed bugs and actionable feature work.
• Use discussions for setup help, ideas, and general community conversation.
• For larger changes, open an issue first so the scope is clear before implementation.
• For security reports, follow SECURITY.md.

Pull Requests

Every PR needs a reason. Your PR description must include:

• what changed and why
• the user or developer impact
• the exact checks you ran
• a linked issue when one exists, using Fixes #123, `Closes `#123, or another clear link
• screenshots when the PR touches UI, terminal presentation, or the VS Code extension
• which provider path was tested when the PR changes provider behavior

The PR author is responsible for ensuring their PR is merge-ready. PRs with merge conflicts will not be reviewed or approved until the conflicts are resolved.

Issues are the recommended starting point for anything non-trivial — opening one first helps avoid wasted effort if the change is out of scope or already being worked on. Small fixes, doc corrections, and obvious improvements can stand on their own without a linked issue, as long as the PR description explains the intent.

What Gets Closed Without Review

PRs may be closed without review...

Files:

• src/query/autoCompactCooldown.test.ts
• src/query.ts
• src/services/compact/autoCompact.test.ts
• src/services/compact/autoCompact.ts

[coderabbitai]

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@src/query/autoCompactCooldown.test.ts`:
- Around line 771-861: The test does not actually verify that
lastForcedFailureAtMs persists through the query loop because both calls to
query() use the same static seedTracking object. To properly validate
persistence, add an onAutoCompactTrackingChange callback to the first query()
call to capture the output tracking state, then pass the captured state (not the
original seedTracking) as the autoCompactTracking parameter to the second
query() call. This ensures the test verifies that the field actually persists
through the query loop's tracking updates rather than simply reusing the same
seed.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Pro Plus

Run ID: 9eb2fd94-4d33-4b6d-b8a5-f41d700a9176

📥 Commits

Reviewing files that changed from the base of the PR and between a8a2321 and 8a8187b.

📒 Files selected for processing (2)

• src/query.ts
• src/query/autoCompactCooldown.test.ts

📜 Review details

🧰 Additional context used

📓 Path-based instructions (5)

**/*.{ts,tsx,js,jsx,py}

📄 CodeRabbit inference engine (CONTRIBUTING.md)

**/*.{ts,tsx,js,jsx,py}: Follow the existing code style in the touched files
Keep comments useful and concise

Files:

• src/query.ts
• src/query/autoCompactCooldown.test.ts

**/*.{ts,tsx}

📄 CodeRabbit inference engine (CONTRIBUTING.md)

Typecheck TypeScript code before submitting (use bun run typecheck)

Files:

• src/query.ts
• src/query/autoCompactCooldown.test.ts

**/*

⚙️ CodeRabbit configuration file

**/*: Apply the OpenClaude maintainer review rubric from AGENTS.md. Review the current diff, not stale discussion context. Separate real blockers from suggestions. Do not request changes for vague style churn. Treat approval as merge-ready from CodeRabbit's side, pending required human review and GitHub Checks. If checks are failing or unavailable, say so clearly instead of implying the PR is fully ready.

Files:

• src/query.ts
• src/query/autoCompactCooldown.test.ts

**

⚙️ CodeRabbit configuration file

**: # Contributing to OpenClaude

Thanks for contributing.

OpenClaude is a fast-moving open-source coding-agent CLI with support for multiple providers, local backends, MCP, and a terminal-first workflow. The best contributions here are focused, well-tested, and easy to review.

Before You Start

• Search existing issues and discussions before opening a new thread.
• Check open pull requests for work that overlaps with your contribution. If a PR already exists that addresses the same change, open an issue or discussion first to align on direction — duplicate PRs may be closed without review.
• Use issues for confirmed bugs and actionable feature work.
• Use discussions for setup help, ideas, and general community conversation.
• For larger changes, open an issue first so the scope is clear before implementation.
• For security reports, follow SECURITY.md.

Pull Requests

Every PR needs a reason. Your PR description must include:

• what changed and why
• the user or developer impact
• the exact checks you ran
• a linked issue when one exists, using Fixes #123, `Closes `#123, or another clear link
• screenshots when the PR touches UI, terminal presentation, or the VS Code extension
• which provider path was tested when the PR changes provider behavior

The PR author is responsible for ensuring their PR is merge-ready. PRs with merge conflicts will not be reviewed or approved until the conflicts are resolved.

Issues are the recommended starting point for anything non-trivial — opening one first helps avoid wasted effort if the change is out of scope or already being worked on. Small fixes, doc corrections, and obvious improvements can stand on their own without a linked issue, as long as the PR description explains the intent.

What Gets Closed Without Review

PRs may be closed without review...

Files:

• src/query.ts
• src/query/autoCompactCooldown.test.ts

{src/**/*.test.ts,src/**/*.test.tsx,tests/**,scripts/**/*.test.ts,vscode-extension/**/*.test.js}

⚙️ CodeRabbit configuration file

{src/**/*.test.ts,src/**/*.test.tsx,tests/**,scripts/**/*.test.ts,vscode-extension/**/*.test.js}: Review tests for meaningful coverage of the changed behavior, isolation of global/env/config state, async cleanup, fake timers, provider profile leaks, and Windows-compatible assumptions. Block when risky runtime changes lack focused regression coverage or tests assert implementation details while missing the user-visible behavior.

Files:

• src/query/autoCompactCooldown.test.ts

🔇 Additional comments (2)

src/query.ts (1)

782-796: LGTM!

src/query/autoCompactCooldown.test.ts (1)

31-32: LGTM!

Also applies to: 48-52

[jatmn]

I found an issue that needs to be addressed before this is ready.

Findings

• [P2] Complete CodeRabbit's request to keep the hard cap active when auto-compact is disabled
  src/services/compact/autoCompact.ts:333
  CodeRabbit's earlier request about this path is still valid on the current head. src/query.ts now sets forceReason: 'message-count' when the new hard cap is exceeded and documents that the cap is always active unless OPENCLAUDE_MAX_ACTIVE_MESSAGES_HARD_CAP=0, but autoCompactIfNeeded() still routes that forced call through shouldAutoCompact(..., true), and this guard returns false whenever DISABLE_AUTO_COMPACT is set or autoCompactEnabled is false. In that configuration an over-cap session proceeds without compacting, so users who disabled token-threshold autocompact lose the new OOM safety net even though they did not disable the hard cap. Please complete the CodeRabbit request by letting the hard-cap forced path bypass the auto-compact-enabled guard, or update the feature contract/tests if the hard cap is intentionally disabled with token autocompact.

[coderabbitai]

Actionable comments posted: 1

Caution

Some comments are outside the diff and can’t be posted inline due to platform limitations.

⚠️ Outside diff range comments (2)

src/query/autoCompactCooldown.test.ts (1)

606-878: ⚠️ Potential issue | 🟠 Major | ⚡ Quick win

The cooldown regressions still miss the low-token / over-cap path.

Every new case here uses overAutoCompactThresholdMessage(), so shouldAutoCompact() stays true and the suite never exercises the hard-cap-only branch where a non-forced skip can clear nextRetryAtMs. That is the branch where the retry storm can still re-arm. Add a many-short-messages scenario that stays below the token threshold across forced failure -> cooled-down skip -> next turn.

As per coding guidelines, "Review tests for meaningful coverage of the changed behavior, isolation of global/env/config state... Block when risky runtime changes lack focused regression coverage."

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/query/autoCompactCooldown.test.ts` around lines 606 - 878, The test suite
for the autocompact cooldown regression only covers scenarios with messages
exceeding the token threshold (using overAutoCompactThresholdMessage()), which
exercises the token-threshold forcing path. Add a new test case that uses short
messages staying below the token threshold to cover the hard-cap-only branch
where a non-forced skip can incorrectly clear nextRetryAtMs during the cooldown
period. This test should verify that when the hard-cap-only condition is active
(no token threshold exceeded) and a cooldown is in effect, the nextRetryAtMs
field is preserved across multiple turns, preventing the retry storm from
re-arming. The test should follow the pattern of the existing three test cases
(hard-cap re-trigger respects a recent forced-failure cool-down, hard-cap
re-trigger fires again once the cool-down has elapsed, and lastForcedFailureAtMs
persists across non-forced skip turns) but use short/normal-sized messages
instead of overAutoCompactThresholdMessage() calls.

Source: Coding guidelines

src/query.ts (1)

618-635: ⚠️ Potential issue | 🟠 Major | ⚡ Quick win

The forced-failure cool-down re-arms once nextRetryAtMs is cleared.

This gate returns false unless both lastForcedFailureAtMs and nextRetryAtMs are present. In the hard-cap case with many short messages, a cooled-down non-forced skip can clear nextRetryAtMs while preserving lastForcedFailureAtMs; the next over-cap turn then restamps forceReason: 'message-count' before the forced-failure cool-down has elapsed, which recreates the retry storm.

Suggested fix

       const forcedCooldownActive = (() => {
         const lastForced = tracking?.lastForcedFailureAtMs
         const nextRetry = tracking?.nextRetryAtMs
-        if (typeof lastForced !== 'number' || typeof nextRetry !== 'number') {
+        if (typeof lastForced !== 'number' || !Number.isFinite(lastForced)) {
           return false
         }
-        if (!Number.isFinite(lastForced) || !Number.isFinite(nextRetry)) {
-          return false
-        }
         const now = Date.now()
-        if (now < nextRetry) {
+        if (
+          typeof nextRetry === 'number' &&
+          Number.isFinite(nextRetry) &&
+          now < nextRetry
+        ) {
           return true
         }
         // Belt-and-suspenders: also honor the cool-down window from
         // the last forced attempt, in case nextRetryAtMs has been
         // cleared by a separate code path.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/query.ts` around lines 618 - 635, The forcedCooldownActive gate requires
both lastForcedFailureAtMs and nextRetryAtMs to be valid numbers before checking
any conditions, but when nextRetryAtMs is cleared by a separate code path, the
function returns false immediately even though lastForcedFailureAtMs may still
indicate an active cool-down. Fix this by allowing the belt-and-suspenders check
(the final return statement comparing now - lastForced against
getAutoCompactFailureCooldownMs()) to work independently when nextRetryAtMs is
not present. Refactor the logic so that the function checks if we're within the
cool-down window based on lastForcedFailureAtMs validity alone, and returns true
if either the nextRetry check or the lastForced cool-down window check succeeds,
rather than requiring both timestamps to be present before performing any
checks.

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@src/services/compact/autoCompact.ts`:
- Around line 439-443: The forced compaction contract is incomplete: the early
return guard checking DISABLE_COMPACT at line 432 of autoCompact.ts still blocks
forced calls before forceReason reaches shouldAutoCompact(), contradicting the
documented behavior. In autoCompact.ts around line 432, modify the
DISABLE_COMPACT guard to only apply for non-forced calls by checking if
forceReason is not set before returning early. In autoCompact.test.ts around
lines 927-1047, add a unit test that verifies forced compaction bypasses the
DISABLE_COMPACT guard. In autoCompactCooldown.test.ts around lines 890-962, add
a corresponding end-to-end regression test with DISABLE_COMPACT=1 to ensure the
hard-cap memory-pressure safety net functions correctly when forced.

---

Outside diff comments:
In `@src/query.ts`:
- Around line 618-635: The forcedCooldownActive gate requires both
lastForcedFailureAtMs and nextRetryAtMs to be valid numbers before checking any
conditions, but when nextRetryAtMs is cleared by a separate code path, the
function returns false immediately even though lastForcedFailureAtMs may still
indicate an active cool-down. Fix this by allowing the belt-and-suspenders check
(the final return statement comparing now - lastForced against
getAutoCompactFailureCooldownMs()) to work independently when nextRetryAtMs is
not present. Refactor the logic so that the function checks if we're within the
cool-down window based on lastForcedFailureAtMs validity alone, and returns true
if either the nextRetry check or the lastForced cool-down window check succeeds,
rather than requiring both timestamps to be present before performing any
checks.

In `@src/query/autoCompactCooldown.test.ts`:
- Around line 606-878: The test suite for the autocompact cooldown regression
only covers scenarios with messages exceeding the token threshold (using
overAutoCompactThresholdMessage()), which exercises the token-threshold forcing
path. Add a new test case that uses short messages staying below the token
threshold to cover the hard-cap-only branch where a non-forced skip can
incorrectly clear nextRetryAtMs during the cooldown period. This test should
verify that when the hard-cap-only condition is active (no token threshold
exceeded) and a cooldown is in effect, the nextRetryAtMs field is preserved
across multiple turns, preventing the retry storm from re-arming. The test
should follow the pattern of the existing three test cases (hard-cap re-trigger
respects a recent forced-failure cool-down, hard-cap re-trigger fires again once
the cool-down has elapsed, and lastForcedFailureAtMs persists across non-forced
skip turns) but use short/normal-sized messages instead of
overAutoCompactThresholdMessage() calls.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Pro Plus

Run ID: 37812e49-c12b-478c-887e-924e1706bbd6

📥 Commits

Reviewing files that changed from the base of the PR and between 655be7f and e40f8a5.

📒 Files selected for processing (4)

• src/query.ts
• src/query/autoCompactCooldown.test.ts
• src/services/compact/autoCompact.test.ts
• src/services/compact/autoCompact.ts

📜 Review details

🧰 Additional context used

📓 Path-based instructions (5)

**/*.{ts,tsx,js,jsx,py}

📄 CodeRabbit inference engine (CONTRIBUTING.md)

**/*.{ts,tsx,js,jsx,py}: Follow the existing code style in the touched files
Keep comments useful and concise

Files:

• src/query.ts
• src/services/compact/autoCompact.test.ts
• src/query/autoCompactCooldown.test.ts
• src/services/compact/autoCompact.ts

**/*.{ts,tsx}

📄 CodeRabbit inference engine (CONTRIBUTING.md)

Typecheck TypeScript code before submitting (use bun run typecheck)

Files:

• src/query.ts
• src/services/compact/autoCompact.test.ts
• src/query/autoCompactCooldown.test.ts
• src/services/compact/autoCompact.ts

**/*

⚙️ CodeRabbit configuration file

**/*: Apply the OpenClaude maintainer review rubric from AGENTS.md. Review the current diff, not stale discussion context. Separate real blockers from suggestions. Do not request changes for vague style churn. Treat approval as merge-ready from CodeRabbit's side, pending required human review and GitHub Checks. If checks are failing or unavailable, say so clearly instead of implying the PR is fully ready.

Files:

• src/query.ts
• src/services/compact/autoCompact.test.ts
• src/query/autoCompactCooldown.test.ts
• src/services/compact/autoCompact.ts

**

⚙️ CodeRabbit configuration file

**: # Contributing to OpenClaude

Thanks for contributing.

OpenClaude is a fast-moving open-source coding-agent CLI with support for multiple providers, local backends, MCP, and a terminal-first workflow. The best contributions here are focused, well-tested, and easy to review.

Before You Start

• Search existing issues and discussions before opening a new thread.
• Check open pull requests for work that overlaps with your contribution. If a PR already exists that addresses the same change, open an issue or discussion first to align on direction — duplicate PRs may be closed without review.
• Use issues for confirmed bugs and actionable feature work.
• Use discussions for setup help, ideas, and general community conversation.
• For larger changes, open an issue first so the scope is clear before implementation.
• For security reports, follow SECURITY.md.

Pull Requests

Every PR needs a reason. Your PR description must include:

• what changed and why
• the user or developer impact
• the exact checks you ran
• a linked issue when one exists, using Fixes #123, `Closes `#123, or another clear link
• screenshots when the PR touches UI, terminal presentation, or the VS Code extension
• which provider path was tested when the PR changes provider behavior

The PR author is responsible for ensuring their PR is merge-ready. PRs with merge conflicts will not be reviewed or approved until the conflicts are resolved.

Issues are the recommended starting point for anything non-trivial — opening one first helps avoid wasted effort if the change is out of scope or already being worked on. Small fixes, doc corrections, and obvious improvements can stand on their own without a linked issue, as long as the PR description explains the intent.

What Gets Closed Without Review

PRs may be closed without review...

Files:

• src/query.ts
• src/services/compact/autoCompact.test.ts
• src/query/autoCompactCooldown.test.ts
• src/services/compact/autoCompact.ts

{src/**/*.test.ts,src/**/*.test.tsx,tests/**,scripts/**/*.test.ts,vscode-extension/**/*.test.js}

⚙️ CodeRabbit configuration file

{src/**/*.test.ts,src/**/*.test.tsx,tests/**,scripts/**/*.test.ts,vscode-extension/**/*.test.js}: Review tests for meaningful coverage of the changed behavior, isolation of global/env/config state, async cleanup, fake timers, provider profile leaks, and Windows-compatible assumptions. Block when risky runtime changes lack focused regression coverage or tests assert implementation details while missing the user-visible behavior.

Files:

• src/services/compact/autoCompact.test.ts
• src/query/autoCompactCooldown.test.ts

[jatmn]

Thanks for the update. I rechecked the changed paths and found an issue that still needs to be addressed.

Findings

• [P2] Keep memory-pressure compaction on the breaker cooldown
  src/query.ts:675
  The new lastForcedFailureAtMs gate only suppresses re-forcing for the message-count cap, but the memory-pressure path below it still unconditionally stamps forceReason: 'memory-pressure' whenever consumeCompactionRequest() returns true. That request is raised repeatedly while RSS remains elevated/critical (src/utils/memoryPressure.ts:128-134), and autoCompactIfNeeded() now lets any forced request bypass the tripped-breaker skip (src/services/compact/autoCompact.ts:500). So if a forced memory-pressure compaction fails, subsequent pressure checks can keep forcing compaction during the 5-minute cooldown instead of honoring the breaker, recreating the retry storm this PR carefully fixed for message-count. Please apply the same forced-failure cooldown gate to memory-pressure requests, or otherwise suppress repeated pressure-forced compactions until the breaker cooldown expires.

[coderabbitai]

Actionable comments posted: 1

Caution

Some comments are outside the diff and can’t be posted inline due to platform limitations.

⚠️ Outside diff range comments (1)

src/query/autoCompactCooldown.test.ts (1)

48-53: ⚠️ Potential issue | 🟡 Minor | ⚡ Quick win

Clear cap env vars in beforeEach to keep test baseline deterministic.

OPENCLAUDE_MAX_ACTIVE_MESSAGES_HARD_CAP and OPENCLAUDE_MAX_ACTIVE_MESSAGES are restored in afterEach, but not reset in beforeEach. Clearing both up front avoids host-environment influence on default-path tests.

Suggested minimal change

   delete process.env.DISABLE_AUTO_COMPACT
   delete process.env.DISABLE_COMPACT
+  delete process.env.OPENCLAUDE_MAX_ACTIVE_MESSAGES_HARD_CAP
+  delete process.env.OPENCLAUDE_MAX_ACTIVE_MESSAGES
   // Some tests below override this to a 60s window to drive the

As per coding guidelines, “Review tests for ... isolation of global/env/config state...”.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/query/autoCompactCooldown.test.ts` around lines 48 - 53, The test file
needs to clear environment variables in the beforeEach hook to ensure test
isolation and deterministic baselines. In addition to the
OPENCLAUDE_AUTOCOMPACT_FAILURE_COOLDOWN_MS cleanup shown in the afterEach block,
add cleanup for OPENCLAUDE_MAX_ACTIVE_MESSAGES_HARD_CAP and
OPENCLAUDE_MAX_ACTIVE_MESSAGES in the beforeEach hook by deleting these
process.env properties. This ensures each test starts with a clean state
unaffected by host environment settings, consistent with the pattern already
established for the cooldown environment variable.

Source: Coding guidelines

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@src/query/autoCompactCooldown.test.ts`:
- Around line 1162-1164: The test mocks the consumeCompactionRequest function
but does not assert that it is actually called, allowing the test to pass
without exercising the memory-pressure request code path. Capture the mocked
consumeCompactionRequest function returned by mock.module in the test setup,
then add an assertion (such as assert.called or equivalent test framework
assertion) to verify that consumeCompactionRequest was invoked during the test
execution. Apply this same assertion pattern at all three locations where this
mock is set up: the anchor location at lines 1162-1164 and the sibling locations
at lines 1185-1186 and 1237-1238.

---

Outside diff comments:
In `@src/query/autoCompactCooldown.test.ts`:
- Around line 48-53: The test file needs to clear environment variables in the
beforeEach hook to ensure test isolation and deterministic baselines. In
addition to the OPENCLAUDE_AUTOCOMPACT_FAILURE_COOLDOWN_MS cleanup shown in the
afterEach block, add cleanup for OPENCLAUDE_MAX_ACTIVE_MESSAGES_HARD_CAP and
OPENCLAUDE_MAX_ACTIVE_MESSAGES in the beforeEach hook by deleting these
process.env properties. This ensures each test starts with a clean state
unaffected by host environment settings, consistent with the pattern already
established for the cooldown environment variable.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Pro Plus

Run ID: 6c9f142f-a8ef-46da-9c94-1aea1b5ffe1d

📥 Commits

Reviewing files that changed from the base of the PR and between 77db19e and 6f35ffd.

📒 Files selected for processing (3)

• src/query.ts
• src/query/autoCompactCooldown.test.ts
• src/utils/memoryPressure.ts

📜 Review details

🧰 Additional context used

📓 Path-based instructions (5)

**/*.{ts,tsx,js,jsx,py}

📄 CodeRabbit inference engine (CONTRIBUTING.md)

**/*.{ts,tsx,js,jsx,py}: Follow the existing code style in the touched files
Keep comments useful and concise

Files:

• src/utils/memoryPressure.ts
• src/query.ts
• src/query/autoCompactCooldown.test.ts

**/*.{ts,tsx}

📄 CodeRabbit inference engine (CONTRIBUTING.md)

Typecheck TypeScript code before submitting (use bun run typecheck)

Files:

• src/utils/memoryPressure.ts
• src/query.ts
• src/query/autoCompactCooldown.test.ts

**/*

⚙️ CodeRabbit configuration file

**/*: Apply the OpenClaude maintainer review rubric from AGENTS.md. Review the current diff, not stale discussion context. Separate real blockers from suggestions. Do not request changes for vague style churn. Treat approval as merge-ready from CodeRabbit's side, pending required human review and GitHub Checks. If checks are failing or unavailable, say so clearly instead of implying the PR is fully ready.

Files:

• src/utils/memoryPressure.ts
• src/query.ts
• src/query/autoCompactCooldown.test.ts

**

⚙️ CodeRabbit configuration file

**: # Contributing to OpenClaude

Thanks for contributing.

OpenClaude is a fast-moving open-source coding-agent CLI with support for multiple providers, local backends, MCP, and a terminal-first workflow. The best contributions here are focused, well-tested, and easy to review.

Before You Start

• Search existing issues and discussions before opening a new thread.
• Check open pull requests for work that overlaps with your contribution. If a PR already exists that addresses the same change, open an issue or discussion first to align on direction — duplicate PRs may be closed without review.
• Use issues for confirmed bugs and actionable feature work.
• Use discussions for setup help, ideas, and general community conversation.
• For larger changes, open an issue first so the scope is clear before implementation.
• For security reports, follow SECURITY.md.

Pull Requests

Every PR needs a reason. Your PR description must include:

• what changed and why
• the user or developer impact
• the exact checks you ran
• a linked issue when one exists, using Fixes #123, `Closes `#123, or another clear link
• screenshots when the PR touches UI, terminal presentation, or the VS Code extension
• which provider path was tested when the PR changes provider behavior

The PR author is responsible for ensuring their PR is merge-ready. PRs with merge conflicts will not be reviewed or approved until the conflicts are resolved.

Issues are the recommended starting point for anything non-trivial — opening one first helps avoid wasted effort if the change is out of scope or already being worked on. Small fixes, doc corrections, and obvious improvements can stand on their own without a linked issue, as long as the PR description explains the intent.

What Gets Closed Without Review

PRs may be closed without review...

Files:

• src/utils/memoryPressure.ts
• src/query.ts
• src/query/autoCompactCooldown.test.ts

{src/**/*.test.ts,src/**/*.test.tsx,tests/**,scripts/**/*.test.ts,vscode-extension/**/*.test.js}

⚙️ CodeRabbit configuration file

{src/**/*.test.ts,src/**/*.test.tsx,tests/**,scripts/**/*.test.ts,vscode-extension/**/*.test.js}: Review tests for meaningful coverage of the changed behavior, isolation of global/env/config state, async cleanup, fake timers, provider profile leaks, and Windows-compatible assumptions. Block when risky runtime changes lack focused regression coverage or tests assert implementation details while missing the user-visible behavior.

Files:

• src/query/autoCompactCooldown.test.ts

🔇 Additional comments (2)

src/query.ts (1)

607-652: LGTM!

Also applies to: 681-697, 793-833

src/utils/memoryPressure.ts (1)

128-134: LGTM!

[jatmn]

I found an issue that needs to be addressed before this is ready.

Findings

• [P1] Complete CodeRabbit's request to remove the duplicate session-memory compact path
  src/commands/compact/compact.ts:84
  CodeRabbit's current review item is still valid: the new session-memory branch calls trySessionMemoryCompaction(...) at line 61 and then falls into the existing progress-wrapped attempt at line 90 instead of returning or closing the branch. The resulting control flow leaves the outer try structurally unbalanced, so bun run build fails with Unexpected catch at this file's outer catch and bun run typecheck reports parse errors in the same file. Please complete that review request by keeping a single session-memory compaction attempt, with the successful cleanup/breaker reset wired into that one path, so /compact builds and cannot compact twice.

• [P2] Keep subagent compactions from changing the main breaker-trip store
  src/services/compact/autoCompact.ts:588
runPostCompactCleanup(querySource) deliberately skips main-thread module-level resets for agent:* sources because subagents share the same process and would otherwise corrupt the main session's state. The new breaker-trip store is also module-level, but the new success paths call clearBreakerTrippedState() unconditionally after runPostCompactCleanup(querySource), and the failure path calls recordBreakerTripped() unconditionally. That means a subagent auto-compact success can clear the main session's "auto-compact paused" signal, or a subagent failure can make the main session look paused, even though no main-session compaction recovered or failed. Please key this state by query source/session, or apply the same main-thread guard used by the cleanup helper before recording or clearing it.

• [P3] Document that off no longer disables every message-count compaction path
  docs/advanced-setup.md:500
  The user-facing docs still say that setting Message-count compaction to off disables the message-count trigger, and the settings summary says "Disabled message-count compaction". This PR adds a default-on hard message-count cap at 1000 messages, and the new tests show it still fires through DISABLE_AUTO_COMPACT, DISABLE_COMPACT, and autoCompactEnabled: false; the only opt-out is OPENCLAUDE_MAX_ACTIVE_MESSAGES_HARD_CAP=0. Please update the docs/settings copy so users understand that off only disables the user-configured threshold, not the operator hard cap.

[jatmn]

Thanks for the update. I rechecked the previously discussed paths and found an issue that still needs to be addressed.

Findings

• [P1] Complete the duplicate session-memory compact cleanup
  src/commands/compact/compact.ts:84
  The current head still leaves the session-memory /compact branch structurally broken: it calls trySessionMemoryCompaction(...) at line 61, enters the success branch without returning or closing that block, and then falls into the progress-wrapped second call at line 90. That keeps the earlier CodeRabbit/maintainer request unresolved and now blocks validation outright: bun run build fails with Unexpected catch at the outer catch, and bun run typecheck reports parse errors in this file. Please keep a single session-memory compaction attempt, wire the breaker reset/cleanup into that successful path, and return before falling through to the traditional compaction path.

• [P2] Key breaker-trip state by session before exposing it
  src/services/compact/compactWarningState.ts:44
  The PR describes the new breaker-trip store as session-scoped and defers REPL/SDK display wiring to a follow-up, but the state added here is a single module-level store with recordBreakerTripped(), getBreakerTripState(), and clearBreakerTrippedState() taking no session key. REPL already runs multiple main-thread sessions in one process and keeps autoCompactTracking in a map keyed by getSessionId(); background sessions also call query() with the same repl_main_thread query source. With the current API, one main-thread session tripping the breaker will make the future UI/API read the trip for every session, and a successful compaction in another main-thread session will clear it. Please key this store by session ID (or thread the existing per-session tracking state through the future reader) before merging the follow-up-facing API.

[coderabbitai]

Actionable comments posted: 2

Caution

Some comments are outside the diff and can’t be posted inline due to platform limitations.

⚠️ Outside diff range comments (1)

src/services/compact/autoCompact.test.ts (1)

1347-1373: ⚠️ Potential issue | 🟡 Minor | ⚡ Quick win

Make breaker-store cleanup unconditional in this describe block.

These tests clear breaker state manually at the end of specific tests; if an assertion fails first, cleanup is skipped and state can leak across tests.

Suggested patch

 describe('subagent (agent:*) compactions do not corrupt the main breaker-trip store', () => {
   beforeEach(() => {
     process.env.CLAUDE_AUTOCOMPACT_PCT_OVERRIDE = '1'
     process.env.OPENCLAUDE_AUTOCOMPACT_FAILURE_COOLDOWN_MS = '5000'
   })
+
+  afterEach(async () => {
+    const { clearBreakerTrippedState } = await import('./compactWarningState.js')
+    clearBreakerTrippedState(getSessionId())
+  })
@@
-    clearBreakerTrippedState(sessionId)
@@
-    clearBreakerTrippedState(sessionId)

As per coding guidelines, {src/**/*.test.ts,src/**/*.test.tsx,tests/**,scripts/**/*.test.ts,vscode-extension/**/*.test.js}: Review tests for meaningful coverage of the changed behavior, isolation of global/env/config state, async cleanup....

Also applies to: 1434-1455

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/services/compact/autoCompact.test.ts` around lines 1347 - 1373, The test
cleanup using clearBreakerTrippedState is currently called at the end of the
test method, which means it will be skipped if an assertion fails beforehand,
causing state to leak across tests. Move the clearBreakerTrippedState call from
the end of this test into an afterEach hook at the describe block level so that
the cleanup executes unconditionally after every test, ensuring breaker state is
properly reset regardless of test outcomes.

Source: Coding guidelines

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@src/services/compact/autoCompact.ts`:
- Around line 2-3: The file imports from the bootstrap/state module using two
different path specifiers: the absolute path 'src/bootstrap/state.js' for
markPostCompaction and the relative path '../../bootstrap/state.js' for
getSdkBetas and getSessionId. Consolidate these into a single import statement
using one consistent path specifier (preferably the relative path to match the
rest of the codebase), combining all imported symbols (markPostCompaction,
getSdkBetas, getSessionId) in a single import statement from bootstrap/state to
ensure they reference the same module instance.

In `@src/services/compact/breakerTripReset.test.ts`:
- Around line 167-207: Replace the dynamic require() calls for
subscribeBreakerTripState with a proper ESM import statement at the top of the
test file to align with ESM guidelines for src/**/*.ts files. Additionally, wrap
the unsubA() and unsubB() cleanup calls in a finally block to ensure
subscriptions are properly unsubscribed and module state is cleared after the
test completes, preventing listener leaks across subsequent test runs where the
module-level breakerTripStores Map would otherwise retain stale subscriptions.

---

Outside diff comments:
In `@src/services/compact/autoCompact.test.ts`:
- Around line 1347-1373: The test cleanup using clearBreakerTrippedState is
currently called at the end of the test method, which means it will be skipped
if an assertion fails beforehand, causing state to leak across tests. Move the
clearBreakerTrippedState call from the end of this test into an afterEach hook
at the describe block level so that the cleanup executes unconditionally after
every test, ensuring breaker state is properly reset regardless of test
outcomes.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Pro Plus

Run ID: 7d278b0e-5e51-4206-a5eb-1c6212701ed2

📥 Commits

Reviewing files that changed from the base of the PR and between 1b40893 and bcabb9b.

📒 Files selected for processing (5)

• src/commands/compact/compact.ts
• src/services/compact/autoCompact.test.ts
• src/services/compact/autoCompact.ts
• src/services/compact/breakerTripReset.test.ts
• src/services/compact/compactWarningState.ts

📜 Review details

🧰 Additional context used

📓 Path-based instructions (13)

src/**/*.{ts,tsx}

📄 CodeRabbit inference engine (AGENTS.md)

Use TypeScript with strict mode and ESM imports

Files:

• src/services/compact/compactWarningState.ts
• src/commands/compact/compact.ts
• src/services/compact/breakerTripReset.test.ts
• src/services/compact/autoCompact.test.ts
• src/services/compact/autoCompact.ts

{src/commands/**/*.ts,src/services/**/*.ts,src/entrypoints/**/*.ts}

📄 CodeRabbit inference engine (AGENTS.md)

Use chalk for terminal color in CLI code

Files:

• src/services/compact/compactWarningState.ts
• src/commands/compact/compact.ts
• src/services/compact/breakerTripReset.test.ts
• src/services/compact/autoCompact.test.ts
• src/services/compact/autoCompact.ts

{src/services/**/*.ts,src/utils/**/*.ts}

📄 CodeRabbit inference engine (AGENTS.md)

Use execa for child processes

Files:

• src/services/compact/compactWarningState.ts
• src/services/compact/breakerTripReset.test.ts
• src/services/compact/autoCompact.test.ts
• src/services/compact/autoCompact.ts

{src/integrations/**/*.ts,src/services/**/*.ts}

📄 CodeRabbit inference engine (AGENTS.md)

Test the exact provider/model path you changed when possible for provider modifications

Files:

• src/services/compact/compactWarningState.ts
• src/services/compact/breakerTripReset.test.ts
• src/services/compact/autoCompact.test.ts
• src/services/compact/autoCompact.ts

**/*.{ts,tsx,js,jsx,py,json,md}

📄 CodeRabbit inference engine (CONTRIBUTING.md)

Follow the existing code style in the touched files

Files:

• src/services/compact/compactWarningState.ts
• src/commands/compact/compact.ts
• src/services/compact/breakerTripReset.test.ts
• src/services/compact/autoCompact.test.ts
• src/services/compact/autoCompact.ts

**/*.{ts,tsx,js,jsx,py}

📄 CodeRabbit inference engine (CONTRIBUTING.md)

Keep comments useful and concise

Files:

• src/services/compact/compactWarningState.ts
• src/commands/compact/compact.ts
• src/services/compact/breakerTripReset.test.ts
• src/services/compact/autoCompact.test.ts
• src/services/compact/autoCompact.ts

**/*.{ts,tsx}

📄 CodeRabbit inference engine (CONTRIBUTING.md)

Follow TypeScript strict mode and type safety practices by running typecheck before submitting

Files:

• src/services/compact/compactWarningState.ts
• src/commands/compact/compact.ts
• src/services/compact/breakerTripReset.test.ts
• src/services/compact/autoCompact.test.ts
• src/services/compact/autoCompact.ts

**/*

⚙️ CodeRabbit configuration file

**/*: Apply the OpenClaude maintainer review rubric from AGENTS.md. Review the current diff, not stale discussion context. Separate real blockers from suggestions. Do not request changes for vague style churn. Treat approval as merge-ready from CodeRabbit's side, pending required human review and GitHub Checks. If checks are failing or unavailable, say so clearly instead of implying the PR is fully ready.

Files:

• src/services/compact/compactWarningState.ts
• src/commands/compact/compact.ts
• src/services/compact/breakerTripReset.test.ts
• src/services/compact/autoCompact.test.ts
• src/services/compact/autoCompact.ts

**

⚙️ CodeRabbit configuration file

**: # Contributing to OpenClaude

Thanks for contributing.

OpenClaude is a fast-moving open-source coding-agent CLI with support for multiple providers, local backends, MCP, and a terminal-first workflow. The best contributions here are focused, well-tested, and easy to review.

Before You Start

• Search existing issues and discussions before opening a new thread.
• Check open pull requests for work that overlaps with your contribution. If a PR already exists that addresses the same change, open an issue or discussion first to align on direction — duplicate PRs may be closed without review.
• Use issues for confirmed bugs and actionable feature work.
• Use discussions for setup help, ideas, and general community conversation.
• For larger changes, open an issue first so the scope is clear before implementation.
• For security reports, follow SECURITY.md.

Pull Requests

Every PR needs a reason. Your PR description must include:

• what changed and why
• the user or developer impact
• the exact checks you ran
• a linked issue when one exists, using Fixes #123, `Closes `#123, or another clear link
• screenshots when the PR touches UI, terminal presentation, or the VS Code extension
• which provider path was tested when the PR changes provider behavior

The PR author is responsible for ensuring their PR is merge-ready. PRs with merge conflicts will not be reviewed or approved until the conflicts are resolved.

Issues are the recommended starting point for anything non-trivial — opening one first helps avoid wasted effort if the change is out of scope or already being worked on. Small fixes, doc corrections, and obvious improvements can stand on their own without a linked issue, as long as the PR description explains the intent.

What Gets Closed Without Review

PRs may be closed without review...

Files:

• src/services/compact/compactWarningState.ts
• src/commands/compact/compact.ts
• src/services/compact/breakerTripReset.test.ts
• src/services/compact/autoCompact.test.ts
• src/services/compact/autoCompact.ts

{src/commands/**/*.ts,src/entrypoints/**/*.ts}

📄 CodeRabbit inference engine (AGENTS.md)

Use commander for CLI argument parsing

Files:

• src/commands/compact/compact.ts

**/*.test.{ts,tsx,js,jsx}

📄 CodeRabbit inference engine (CONTRIBUTING.md)

Add or update tests when the change affects behavior

Files:

• src/services/compact/breakerTripReset.test.ts
• src/services/compact/autoCompact.test.ts

**/*.test.{ts,tsx,js}

📄 CodeRabbit inference engine (CONTRIBUTING.md)

Test the exact provider/model path you changed when possible

Files:

• src/services/compact/breakerTripReset.test.ts
• src/services/compact/autoCompact.test.ts

{src/**/*.test.ts,src/**/*.test.tsx,tests/**,scripts/**/*.test.ts,vscode-extension/**/*.test.js}

⚙️ CodeRabbit configuration file

{src/**/*.test.ts,src/**/*.test.tsx,tests/**,scripts/**/*.test.ts,vscode-extension/**/*.test.js}: Review tests for meaningful coverage of the changed behavior, isolation of global/env/config state, async cleanup, fake timers, provider profile leaks, and Windows-compatible assumptions. Block when risky runtime changes lack focused regression coverage or tests assert implementation details while missing the user-visible behavior.

Files:

• src/services/compact/breakerTripReset.test.ts
• src/services/compact/autoCompact.test.ts

🔇 Additional comments (2)

src/services/compact/compactWarningState.ts (1)

21-35: LGTM!

Also applies to: 54-75, 78-144

src/commands/compact/compact.ts (1)

3-3: LGTM!

Also applies to: 76-80, 143-143, 230-230

[jatmn]

I found a runtime gap that still leaves part of the reported OOM path unbounded, plus one test cleanup item from the existing review thread.

Findings

• [P2] Prune teammate history when hard-cap compaction yields a boundary
  src/utils/swarm/inProcessRunner.ts:1244
  Issue #1373 also calls out the in-process teammate allMessages buffer as an unbounded accumulator. This PR adds the hard cap inside query.ts, and runAgent records/yields system compact_boundary messages from that query path, but the teammate runner still appends every yielded message to its lifetime allMessages array without applying the same compact-boundary pruning that QueryEngine does. As a result, a teammate can trigger the new hard-cap compaction inside query() while the runner keeps the entire pre-compact prefix, including full tool results, alive in allMessages until a separate token-threshold compaction happens. Please prune allMessages/task state when a compact boundary comes back from runAgent, or route this path through the same post-boundary replacement logic used by the explicit teammate compaction branch.

• [P3] Complete CodeRabbit's breaker-trip test cleanup requests
  src/services/compact/breakerTripReset.test.ts:171
src/services/compact/autoCompact.test.ts:1372
  CodeRabbit's cleanup request is still only partially addressed. The subscription test registers unsubA/unsubB but only calls them after all assertions complete, so any assertion failure before the bottom of the test leaves those module-level breaker-trip listeners registered for the rest of the process. The subagent breaker tests also still seed the main session's breaker state and clear it only at the end of the test body, so an early failure can leave that state behind. Please complete that review request by moving the breaker-state cleanup into the describe-level afterEach and wrapping the subscription test's unsubscribe calls in a finally block.

[coderabbitai]

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@src/utils/swarm/inProcessRunner.ts`:
- Around line 1243-1249: The compact_boundary message handling in the block
checking message.subtype === 'compact_boundary' is using the wrong field name
and performing a destructive splice even when metadata is absent. Change
message.compactMetadata?.preservedPrefix to
message.compactMetadata?.preservedSegment to align with the compact-boundary
contract used in related code, and add a guard condition to only perform the
splice operation when preservedSegment is actually present, preventing the loss
of valid message history when the metadata is missing.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Pro Plus

Run ID: 0cf3ab77-c67a-466a-8368-8b0fd35b14ab

📥 Commits

Reviewing files that changed from the base of the PR and between a95fe57 and c824dc2.

📒 Files selected for processing (3)

• src/services/compact/autoCompact.test.ts
• src/services/compact/breakerTripReset.test.ts
• src/utils/swarm/inProcessRunner.ts

📜 Review details

🧰 Additional context used

📓 Path-based instructions (12)

src/**/*.{ts,tsx}

📄 CodeRabbit inference engine (AGENTS.md)

Use TypeScript with strict mode and ESM imports

Files:

• src/utils/swarm/inProcessRunner.ts
• src/services/compact/breakerTripReset.test.ts
• src/services/compact/autoCompact.test.ts

{src/services/**/*.ts,src/utils/**/*.ts}

📄 CodeRabbit inference engine (AGENTS.md)

Use execa for child processes

Files:

• src/utils/swarm/inProcessRunner.ts
• src/services/compact/breakerTripReset.test.ts
• src/services/compact/autoCompact.test.ts

**/*.{ts,tsx,js,jsx,py,json,md}

📄 CodeRabbit inference engine (CONTRIBUTING.md)

Follow the existing code style in the touched files

Files:

• src/utils/swarm/inProcessRunner.ts
• src/services/compact/breakerTripReset.test.ts
• src/services/compact/autoCompact.test.ts

**/*.{ts,tsx,js,jsx,py}

📄 CodeRabbit inference engine (CONTRIBUTING.md)

Keep comments useful and concise

Files:

• src/utils/swarm/inProcessRunner.ts
• src/services/compact/breakerTripReset.test.ts
• src/services/compact/autoCompact.test.ts

**/*.{ts,tsx}

📄 CodeRabbit inference engine (CONTRIBUTING.md)

Follow TypeScript strict mode and type safety practices by running typecheck before submitting

Files:

• src/utils/swarm/inProcessRunner.ts
• src/services/compact/breakerTripReset.test.ts
• src/services/compact/autoCompact.test.ts

**/*

⚙️ CodeRabbit configuration file

**/*: Apply the OpenClaude maintainer review rubric from AGENTS.md. Review the current diff, not stale discussion context. Separate real blockers from suggestions. Do not request changes for vague style churn. Treat approval as merge-ready from CodeRabbit's side, pending required human review and GitHub Checks. If checks are failing or unavailable, say so clearly instead of implying the PR is fully ready.

Files:

• src/utils/swarm/inProcessRunner.ts
• src/services/compact/breakerTripReset.test.ts
• src/services/compact/autoCompact.test.ts

**

⚙️ CodeRabbit configuration file

**: # Contributing to OpenClaude

Thanks for contributing.

OpenClaude is a fast-moving open-source coding-agent CLI with support for multiple providers, local backends, MCP, and a terminal-first workflow. The best contributions here are focused, well-tested, and easy to review.

Before You Start

• Search existing issues and discussions before opening a new thread.
• Check open pull requests for work that overlaps with your contribution. If a PR already exists that addresses the same change, open an issue or discussion first to align on direction — duplicate PRs may be closed without review.
• Use issues for confirmed bugs and actionable feature work.
• Use discussions for setup help, ideas, and general community conversation.
• For larger changes, open an issue first so the scope is clear before implementation.
• For security reports, follow SECURITY.md.

Pull Requests

Every PR needs a reason. Your PR description must include:

• what changed and why
• the user or developer impact
• the exact checks you ran
• a linked issue when one exists, using Fixes #123, `Closes `#123, or another clear link
• screenshots when the PR touches UI, terminal presentation, or the VS Code extension
• which provider path was tested when the PR changes provider behavior

The PR author is responsible for ensuring their PR is merge-ready. PRs with merge conflicts will not be reviewed or approved until the conflicts are resolved.

Issues are the recommended starting point for anything non-trivial — opening one first helps avoid wasted effort if the change is out of scope or already being worked on. Small fixes, doc corrections, and obvious improvements can stand on their own without a linked issue, as long as the PR description explains the intent.

What Gets Closed Without Review

PRs may be closed without review...

Files:

• src/utils/swarm/inProcessRunner.ts
• src/services/compact/breakerTripReset.test.ts
• src/services/compact/autoCompact.test.ts

{src/commands/**/*.ts,src/services/**/*.ts,src/entrypoints/**/*.ts}

📄 CodeRabbit inference engine (AGENTS.md)

Use chalk for terminal color in CLI code

Files:

• src/services/compact/breakerTripReset.test.ts
• src/services/compact/autoCompact.test.ts

{src/integrations/**/*.ts,src/services/**/*.ts}

📄 CodeRabbit inference engine (AGENTS.md)

Test the exact provider/model path you changed when possible for provider modifications

Files:

• src/services/compact/breakerTripReset.test.ts
• src/services/compact/autoCompact.test.ts

**/*.test.{ts,tsx,js,jsx}

📄 CodeRabbit inference engine (CONTRIBUTING.md)

Add or update tests when the change affects behavior

Files:

• src/services/compact/breakerTripReset.test.ts
• src/services/compact/autoCompact.test.ts

**/*.test.{ts,tsx,js}

📄 CodeRabbit inference engine (CONTRIBUTING.md)

Test the exact provider/model path you changed when possible

Files:

• src/services/compact/breakerTripReset.test.ts
• src/services/compact/autoCompact.test.ts

{src/**/*.test.ts,src/**/*.test.tsx,tests/**,scripts/**/*.test.ts,vscode-extension/**/*.test.js}

⚙️ CodeRabbit configuration file

{src/**/*.test.ts,src/**/*.test.tsx,tests/**,scripts/**/*.test.ts,vscode-extension/**/*.test.js}: Review tests for meaningful coverage of the changed behavior, isolation of global/env/config state, async cleanup, fake timers, provider profile leaks, and Windows-compatible assumptions. Block when risky runtime changes lack focused regression coverage or tests assert implementation details while missing the user-visible behavior.

Files:

• src/services/compact/breakerTripReset.test.ts
• src/services/compact/autoCompact.test.ts

🔇 Additional comments (5)

src/services/compact/breakerTripReset.test.ts (1)

175-183: Dynamic require() still violates ESM guideline.

The subscription leak is now fixed with try/finally cleanup—nice. However, the dynamic require('./compactWarningState.js') at lines 177 and 181 still contradicts the ESM import guideline for src/**/*.ts files. Add subscribeBreakerTripState to the static import at line 7–11 and call it directly.

Source: Coding guidelines

src/services/compact/autoCompact.test.ts (4)

31-32: LGTM!

---

1350-1377: LGTM!

---

1395-1432: LGTM!

---

1449-1471: LGTM!

[coderabbitai]

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@src/utils/swarm/inProcessRunner.ts`:
- Around line 1243-1249: The issue is in the inProcessRunner.ts file where the
code attempts to spread preservedSegment as if it were an array of messages, but
preservedSegment is actually a metadata object and not iterable, which causes a
runtime crash. In the compact_boundary message handler where allMessages.splice
is called, remove the spread operator and instead handle preservedSegment as the
object metadata it is. Verify what the correct property or structure is that
contains the actual messages array to prune, and either extract that array from
within the preservedSegment object or reassign it appropriately based on the
actual data structure returned by the QueryEngine's compaction logic.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Pro Plus

Run ID: caaefc11-8fac-421f-85f6-5b6de5919877

📥 Commits

Reviewing files that changed from the base of the PR and between c824dc2 and 0353f80.

📒 Files selected for processing (3)

• src/services/compact/autoCompact.test.ts
• src/services/compact/breakerTripReset.test.ts
• src/utils/swarm/inProcessRunner.ts

📜 Review details

🧰 Additional context used

📓 Path-based instructions (12)

src/**/*.{ts,tsx}

📄 CodeRabbit inference engine (AGENTS.md)

Use TypeScript with strict mode and ESM imports

Files:

• src/utils/swarm/inProcessRunner.ts
• src/services/compact/breakerTripReset.test.ts
• src/services/compact/autoCompact.test.ts

{src/services/**/*.ts,src/utils/**/*.ts}

📄 CodeRabbit inference engine (AGENTS.md)

Use execa for child processes

Files:

• src/utils/swarm/inProcessRunner.ts
• src/services/compact/breakerTripReset.test.ts
• src/services/compact/autoCompact.test.ts

**/*.{ts,tsx,js,jsx,py,json,md}

📄 CodeRabbit inference engine (CONTRIBUTING.md)

Follow the existing code style in the touched files

Files:

• src/utils/swarm/inProcessRunner.ts
• src/services/compact/breakerTripReset.test.ts
• src/services/compact/autoCompact.test.ts

**/*.{ts,tsx,js,jsx,py}

📄 CodeRabbit inference engine (CONTRIBUTING.md)

Keep comments useful and concise

Files:

• src/utils/swarm/inProcessRunner.ts
• src/services/compact/breakerTripReset.test.ts
• src/services/compact/autoCompact.test.ts

**/*.{ts,tsx}

📄 CodeRabbit inference engine (CONTRIBUTING.md)

Follow TypeScript strict mode and type safety practices by running typecheck before submitting

Files:

• src/utils/swarm/inProcessRunner.ts
• src/services/compact/breakerTripReset.test.ts
• src/services/compact/autoCompact.test.ts

**/*

⚙️ CodeRabbit configuration file

**/*: Apply the OpenClaude maintainer review rubric from AGENTS.md. Review the current diff, not stale discussion context. Separate real blockers from suggestions. Do not request changes for vague style churn. Treat approval as merge-ready from CodeRabbit's side, pending required human review and GitHub Checks. If checks are failing or unavailable, say so clearly instead of implying the PR is fully ready.

Files:

• src/utils/swarm/inProcessRunner.ts
• src/services/compact/breakerTripReset.test.ts
• src/services/compact/autoCompact.test.ts

**

⚙️ CodeRabbit configuration file

**: # AGENTS.md - AI Agent Coding Guide

This guide is for AI coding agents working in the OpenClaude repository. Read it before changing code, and also follow CONTRIBUTING.md for contributor policy, PR expectations, review follow-up, and project scope.

Project Snapshot

OpenClaude is a coding-agent CLI for cloud and local model providers. It supports OpenAI-compatible APIs, Anthropic, Gemini, DeepSeek, Ollama, MCP, local backends, slash commands, tools, agents, and a React/Ink terminal UI.

The installed CLI runs on Node.js >=22.0.0. Bun is used for source builds, scripts, dependency management, and tests.

Work Style

• Keep changes focused on one problem.
• Prefer existing patterns in the file or nearby module.
• Avoid unrelated formatting, renames, dependency changes, or broad rewrites.
• Add or update tests when behavior changes.
• Update docs when setup, commands, provider behavior, or user-facing behavior changes.
• For new features, larger refactors, dependencies, or runtime changes, follow the issue-first guidance in CONTRIBUTING.md.

Stack And Conventions

• TypeScript with strict mode and ESM imports.
• React + Ink for terminal UI.
• Bun lockfile and Bun scripts for development workflows.
• Node runtime for the built CLI.
• Python exists for legacy/local-provider helper code. Do not add new Python code or expand Python-based features unless a maintainer explicitly approves that direction.

Common libraries and patterns:

• chalk for terminal color.
• commander for CLI argument parsing.
• execa for child processes.
• Existing service, provider, settings, permission, and UI patterns over new abstractions.

Repository Map

• src/commands/ - slash and CLI command implementations.
• src/components/ - React/Ink UI components.
• src/services/ - API, MCP, OAuth, wiki, voice, and other service integrations.
• src/tools/ - tool implementations.
• src/utils/ - shared utilities.
• `src/integration...

Files:

• src/utils/swarm/inProcessRunner.ts
• src/services/compact/breakerTripReset.test.ts
• src/services/compact/autoCompact.test.ts

{src/commands/**/*.ts,src/services/**/*.ts,src/entrypoints/**/*.ts}

📄 CodeRabbit inference engine (AGENTS.md)

Use chalk for terminal color in CLI code

Files:

• src/services/compact/breakerTripReset.test.ts
• src/services/compact/autoCompact.test.ts

{src/integrations/**/*.ts,src/services/**/*.ts}

📄 CodeRabbit inference engine (AGENTS.md)

Test the exact provider/model path you changed when possible for provider modifications

Files:

• src/services/compact/breakerTripReset.test.ts
• src/services/compact/autoCompact.test.ts

**/*.test.{ts,tsx,js,jsx}

📄 CodeRabbit inference engine (CONTRIBUTING.md)

Add or update tests when the change affects behavior

Files:

• src/services/compact/breakerTripReset.test.ts
• src/services/compact/autoCompact.test.ts

**/*.test.{ts,tsx,js}

📄 CodeRabbit inference engine (CONTRIBUTING.md)

Test the exact provider/model path you changed when possible

Files:

• src/services/compact/breakerTripReset.test.ts
• src/services/compact/autoCompact.test.ts

{src/**/*.test.ts,src/**/*.test.tsx,tests/**,scripts/**/*.test.ts,vscode-extension/**/*.test.js}

⚙️ CodeRabbit configuration file

{src/**/*.test.ts,src/**/*.test.tsx,tests/**,scripts/**/*.test.ts,vscode-extension/**/*.test.js}: Review tests for meaningful coverage of the changed behavior, isolation of global/env/config state, async cleanup, fake timers, provider profile leaks, and Windows-compatible assumptions. Block when risky runtime changes lack focused regression coverage or tests assert implementation details while missing the user-visible behavior.

Files:

• src/services/compact/breakerTripReset.test.ts
• src/services/compact/autoCompact.test.ts

🔇 Additional comments (7)

src/services/compact/breakerTripReset.test.ts (3)

175-183: ⚡ Quick win

Dynamic require() still violates ESM guidelines.

The finally block from the previous review was added—nice. But the dynamic require() calls remain. The comment claims "hoisting issues," but subscribeBreakerTripState is a plain export like clearBreakerTrippedState/getBreakerTripState/recordBreakerTripped, all of which are imported statically at the top of this file without issue.

Add subscribeBreakerTripState to the existing static import and remove the inline require() calls.

Source: Coding guidelines

---

1-166: LGTM!

---

214-218: LGTM!

src/services/compact/autoCompact.test.ts (4)

31-33: LGTM!

---

1350-1377: LGTM!

---

1395-1432: LGTM!

---

1449-1471: LGTM!

[jatmn]

I found a couple of issues that need to be addressed before this is ready.

Findings

• [P2] Prune teammate history for every compact boundary
  src/utils/swarm/inProcessRunner.ts:1244
  This only prunes allMessages when the yielded compact_boundary has compactMetadata.preservedSegment, but the normal auto-compact path through compactConversation() returns an unannotated boundary (no messagesToKeep / preserved segment). In that common hard-cap/full-compact case, the teammate keeps the entire pre-compact allMessages buffer, appends the boundary and summary, and then passes the bloated history back as forkContextMessages on the next iteration. That means the secondary OOM path from issue #1373 is still not actually bounded for regular teammate auto-compactions. Please reset the buffer on every compact boundary to mirror the post-compact message stream, not only the preserved-segment variant.

• [P3] Complete CodeRabbit's request to use the static ESM import
  src/services/compact/breakerTripReset.test.ts:177
  CodeRabbit's earlier review item asked for subscribeBreakerTripState to be added to the existing module-scope import, but the current patch still dynamically calls require('./compactWarningState.js') in this TypeScript test. Please complete that review request by importing subscribeBreakerTripState alongside clearBreakerTrippedState, getBreakerTripState, and recordBreakerTripped, then call it directly in the session-isolation test.

[jatmn]

Thanks for the update. I found one issue that still needs to be addressed.

Findings

• [P2] Block over-cap low-token histories while forced compaction cools down
  src/query.ts:979
  The new hard-cap cooldown gate can defer forceReason: 'hard-message-count', but the follow-up blocking check still only considers token pressure (isAboveAutoCompactThreshold / tokenUsage >= getAutoCompactThreshold(model)). That leaves the exact low-token/high-message-count path this cap is meant to bound free to continue calling the model during cooldown after repeated forced failures. The new regression test demonstrates the problem: with two short messages, hard cap 1, a recent forced failure, and a tripped breaker, the test expects terminal.reason to be max_turns rather than blocking_limit (src/query/autoCompactCooldown.test.ts:1136). In a real long-running session with many small tool-result turns, this means state.messages can keep growing while the provider is down, so the “absolute cap” no longer prevents the #1373 runaway path. Please include the active hard-cap-overrun/forced-cooldown state in the blocking condition, or otherwise stop before sending another over-cap request when the safety-net compaction is cooling down.

[coderabbitai]

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@src/query.ts`:
- Around line 983-988: In the isInForcedCooldown constant definition, change the
field reference from `tracking?.lastFailureAtMs` to
`tracking?.lastForcedFailureAtMs`. These are distinct fields where
lastFailureAtMs tracks natural cooldown while lastForcedFailureAtMs tracks
forced-attempt cooldown. Using the wrong field breaks the forced-cooldown
blocking logic in the isInMessageCountCooldown check that follows, which
prevents retry storms on the message-count blocking path.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Pro Plus

Run ID: d736696b-8921-4395-a634-cc51a07b9a8d

📥 Commits

Reviewing files that changed from the base of the PR and between afecc4c and caf86b4.

📒 Files selected for processing (1)

• src/query.ts

📜 Review details

🧰 Additional context used

📓 Path-based instructions (6)

src/**/*.{ts,tsx}

📄 CodeRabbit inference engine (AGENTS.md)

Use TypeScript with strict mode and ESM imports

Files:

• src/query.ts

**/*.{ts,tsx,js,jsx,py,json,md}

📄 CodeRabbit inference engine (CONTRIBUTING.md)

Follow the existing code style in the touched files

Files:

• src/query.ts

**/*.{ts,tsx,js,jsx,py}

📄 CodeRabbit inference engine (CONTRIBUTING.md)

Keep comments useful and concise

Files:

• src/query.ts

**/*.{ts,tsx}

📄 CodeRabbit inference engine (CONTRIBUTING.md)

Follow TypeScript strict mode and type safety practices by running typecheck before submitting

Files:

• src/query.ts

**/*

⚙️ CodeRabbit configuration file

**/*: Apply the OpenClaude maintainer review rubric from AGENTS.md. Review the current diff, not stale discussion context. Separate real blockers from suggestions. Do not request changes for vague style churn. Treat approval as merge-ready from CodeRabbit's side, pending required human review and GitHub Checks. If checks are failing or unavailable, say so clearly instead of implying the PR is fully ready.

Files:

• src/query.ts

**

⚙️ CodeRabbit configuration file

**: # AGENTS.md - AI Agent Coding Guide

This guide is for AI coding agents working in the OpenClaude repository. Read it before changing code, and also follow CONTRIBUTING.md for contributor policy, PR expectations, review follow-up, and project scope.

Project Snapshot

OpenClaude is a coding-agent CLI for cloud and local model providers. It supports OpenAI-compatible APIs, Anthropic, Gemini, DeepSeek, Ollama, MCP, local backends, slash commands, tools, agents, and a React/Ink terminal UI.

The installed CLI runs on Node.js >=22.0.0. Bun is used for source builds, scripts, dependency management, and tests.

Work Style

• Keep changes focused on one problem.
• Prefer existing patterns in the file or nearby module.
• Avoid unrelated formatting, renames, dependency changes, or broad rewrites.
• Add or update tests when behavior changes.
• Update docs when setup, commands, provider behavior, or user-facing behavior changes.
• For new features, larger refactors, dependencies, or runtime changes, follow the issue-first guidance in CONTRIBUTING.md.

Stack And Conventions

• TypeScript with strict mode and ESM imports.
• React + Ink for terminal UI.
• Bun lockfile and Bun scripts for development workflows.
• Node runtime for the built CLI.
• Python exists for legacy/local-provider helper code. Do not add new Python code or expand Python-based features unless a maintainer explicitly approves that direction.

Common libraries and patterns:

• chalk for terminal color.
• commander for CLI argument parsing.
• execa for child processes.
• Existing service, provider, settings, permission, and UI patterns over new abstractions.

Repository Map

• src/commands/ - slash and CLI command implementations.
• src/components/ - React/Ink UI components.
• src/services/ - API, MCP, OAuth, wiki, voice, and other service integrations.
• src/tools/ - tool implementations.
• src/utils/ - shared utilities.
• `src/integration...

Files:

• src/query.ts

[jatmn]

Thanks for the update. I found issues that still need to be addressed.

Findings

• [P2] Complete the over-cap cooldown block for low-token histories
  src/query.ts:983
  The current forced-cooldown block still depends on tracking.forceReason, but the real autoCompactIfNeeded() path consumes that flag before returning (src/services/compact/autoCompact.ts:450-455), and the cool-down deferral branch intentionally does not re-stamp it. That means after a forced hard-cap failure, the next low-token/high-message-count turn has lastForcedFailureAtMs and nextRetryAtMs but no forceReason, so isInForcedCooldown is false and isAboveBreakerThreshold is also false for small messages. The new regression test captures the bad behavior by expecting terminal.reason to be max_turns for two short messages over a hard cap of 1 while forced cooldown is active (src/query/autoCompactCooldown.test.ts:1059-1145). In production that still sends another model request while state.messages is over the absolute cap and compaction is cooling down, so the safety net does not prevent the #1373 runaway path for many small tool-result turns. Please complete the earlier review request by deriving the blocking condition from the over-active-cap plus forced-cooldown state, not from the one-shot forceReason flag.

• [P3] Clear breaker-trip state after partial compact
  src/screens/REPL.tsx:5047
  This PR moved breaker-trip clearing out of runPostCompactCleanup(), so every successful compaction path now has to clear the new compactWarningState breaker store explicitly. The message-selector partial compact path clears the old autoCompactTracking state (setAutoCompactTrackingForSession(..., undefined)), then calls runPostCompactCleanup(), but it never calls clearBreakerTrippedState(getSessionId()). After a user successfully summarizes via that path, the newly exposed breaker-trip state can remain stuck on "tripped" even though the session just recovered. Please clear the breaker-trip store alongside the existing tracking reset in this successful partial-compact path.