fix(cc): CLAUDE_CODE_DISABLE_AGENT_VIEW=1 in provider_manifest env - the one documented off-switch for Claude Code 2.1.258's background daemon (DPLAN-0324, Patrick's word 23:25, manual BAUD test passed 23:42/23:53). THE INCIDENT: three devpulse sessions alive at once, resumable from the phone, while Ctrl-C x2 could not restart the seat. Not adoption - a human hit option 2 (Move to background and exit) at the new exit dialog once, and from then on BAUD's -c door SKIPPED the live background job (2.1.257 no-longer-reopens-a-stalled-copy) and landed on the next transcript down, resurrecting a stopped Aug-31 fork four times in one night. THE FIX IS A RESTORE, NOT A REDESIGN (Patrick: its not a ruling, its how baud worked - rewire for the exact same results): disableAgentView / CLAUDE_CODE_DISABLE_AGENT_VIEW turns off claude agents, --bg, /background, the on-demand supervisor AND removes option 2 from the dialog = 2.1.228 behaviour back with BAUD, session_boot and presence_gate untouched. Cost: /fork old behaviour and /subtask, neither used here. Shipped through the manifest env block because that is the channel aipass doctor renders and checks against ~/.claude/settings.json (doctor: 35 pass, 0 errors once user settings carried it). MEASURED, not assumed, with throwaway probes and no pytest (test-session work is incomplete by Patrick's rule): user-level settings gate every subcommand (agents/stop/attach/daemon/respawn/--bg all answer disabled); project-level .claude/settings.json does NOT gate the CLI, so that edit was reverted; settings env 1 beats a shell =0, so claude stop is unreachable once the switch is on - the switch was lifted for one minute so the manual sequence (claude stop 98c9785a, then BAUD resume) could run, then re-applied; a SIGTERMed bg job is NOT respawned by the daemon (hooks' cc_sessions.py:317 docstring says it is - unsourced, on the hardening list); drone @hooks sessions reclaim refuses bg kinds, so my first sequence would have stopped nothing - caught by the pointer-graph report before Patrick ran it. RESULT: one devpulse, interactive in tmux, Fable 5.1, zero bg jobs, daemon and warm spares gone, Ctrl-C x2 exits clean, BAUD agent-card resume lands in the same chat. Hardening (plan only, owners mailed next): presence_gate bg exemption (ruling a, DPLAN-0310) now rests on a false premise; session_boot daemon-stop paths dead code; manifest pinned_version 2.1.228 vs running 2.1.258; phone and desktop doors disagree for a bg seat (pointer_graph.md sec 5.3); who may hold the pointer pen for managers is open in the record (sec 6).

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Hw2gQBYHFHLnyYPCT7UC8C

Author: AIOSAI

.claude/provider_manifest.json

@@ -147,7 +147,8 @@
       ],
       "env": {
         "AIPASS_HOME": "{{REPO_ROOT}}",
-        "CLAUDE_CODE_DISABLE_AUTO_MEMORY": "1"
+        "CLAUDE_CODE_DISABLE_AUTO_MEMORY": "1",
+        "CLAUDE_CODE_DISABLE_AGENT_VIEW": "1"
       },
       "permissions": {
         "deny": [