Autonomi (Loki Mode) now consumes OpenSpec for autonomous multi-agent execution

[asklokesh]

We built native OpenSpec integration into Loki Mode (v6.11.0). You can now plan with OpenSpec and execute with parallel autonomous agents.

What it does

# Plan with OpenSpec as usual
/opsx:propose add-auth
/opsx:ff

# Hand off to Loki Mode for autonomous execution
loki start --openspec ./openspec/changes/add-auth/

Loki Mode reads your proposal.md, specs/, design.md, and tasks.md. It parses your delta specs (ADDED/MODIFIED/REMOVED), maps tasks to its agent queue, and executes with RARV (Reason-Act-Reflect-Verify) cycles and quality gates. No manual translation step.

Why delta specs matter here

The delta spec format is the real value. When agents receive structured MODIFIED context, they update existing code in place instead of creating duplicate files. REMOVED context triggers deprecation, not orphaned code. This is something we struggled with before — autonomous agents are bad at brownfield work without explicit change context. OpenSpec's delta format solves that directly.

How it works under the hood

The adapter (openspec-adapter.py, stdlib-only Python, no dependency on @fission-ai/openspec) does:

1. Validates the OpenSpec change directory structure
2. Parses each artifact into structured data
3. Extracts delta context from spec files
4. Generates a normalized PRD that Loki Mode's RARV cycle consumes
5. Injects delta context so agents know what's changing vs what's new
6. Maps tasks.md checkboxes directly to the execution queue (already-checked items are skipped)

The flag is mutually exclusive with --bmad-project (our other structured input adapter). Non-OpenSpec users are completely unaffected.

What shipped in v6.11.0

• autonomy/openspec-adapter.py — 827-line stdlib-only adapter
• skills/openspec-integration.md — delta-aware agent skill module
• 28 unit tests, 6 example fixture directories
• CLI: --openspec PATH flag on both loki start and autonomy/run.sh

One honest thing

During E2E testing we found a bug where the CLI adapter ran validation only without generating output files, so the normalized PRD was never created and execution silently fell through. Fixed before release, but worth mentioning because "it works on my machine" is the enemy of trust.

What's next

We'd be interested in feedback from OpenSpec users who try this. Specifically:

• Does the adapter handle your spec format correctly? We tested against the documented format and 6 fixture variations, but real-world usage always surfaces edge cases.
• Would a custom OpenSpec schema optimized for Loki Mode execution (openspec/schemas/loki-mode/) be useful? It would generate tasks.md with parallelization hints and complexity markers.
• Is there interest in bidirectional sync where Loki Mode marks tasks.md checkboxes as it completes them?

GitHub: https://github.com/asklokesh/loki-mode
Website: https://www.autonomi.dev/
Docs: https://www.autonomi.dev/docs/openspec-integration
Discord: https://discord.gg/k8NpBhc5KA

---