[broven]

Summary

• recover no longer reports healthy while a client outside the session's executor process is the attacher of the session's tab. Before this change, that attacher refused every fresh attach (already attached by another client), recover returned straight from the state machine's healthy without checking, and session reset (which only reaps the executor) could not free it.
• Clients now report their pid on the control ws (?pid=). recover gains a rung 0, DaemonState.release_foreign_attachers, which releases attacher ownership of the session's tabs held outside the live executor process. Each release is reported as an attach step. The resident executor's own attachment is kept, as designed.
• The attach-conflict fix hint now says what each verb can free: recover releases an outside attacher; if the error survives a healthy recover, session reset frees an attacher inside the executor process. ADR-0013 is amended.
• Known limits (also in the commit message):
  • Matching is per process, so an extra attacher inside the executor process still needs session reset.
  • A client too old to report its pid is treated as outside.
  • Rung 0 applies to every backend, per ADR-0013's backend-agnostic ladder.
  • The incident's trigger (PageBindTimeout, and connections that kept piling up per call) was not reproduced and is out of scope.

Closes #131

Test evidence

New E2E test, tests/daemon/e2e/test_issue131_stray_attacher.py (real Chrome for Testing + unpacked extension):

• A stray process attaches the session's tab and stays alive for the whole test. A fresh attach is refused. recover reports the release, and after session reset a new outside process can attach.
  • It was red before the fix: recover cold-started the executor and released nothing.
• Negative control: when the resident executor is the attacher, recover releases nothing.

bash tests/daemon/e2e/run.sh (full real-Chrome E2E)

tests/daemon/e2e/test_l3_parity.py ...                                   [ 78%]
tests/daemon/e2e/test_long_lived_session_degradation.py .                [ 79%]
tests/daemon/e2e/test_markdown_command_extension.py ...                  [ 82%]
tests/daemon/e2e/test_markdown_normalize_real_dom.py ...                 [ 85%]
tests/daemon/e2e/test_mute_session_groups.py .                           [ 86%]
tests/daemon/e2e/test_patch_extension.py ..                              [ 88%]
tests/daemon/e2e/test_repro_extension_reload_orphan.py ....              [ 92%]
tests/daemon/e2e/test_repro_zer123.py .                                  [ 93%]
tests/daemon/e2e/test_search_command_cdp.py ....                         [ 97%]
tests/daemon/e2e/test_userscripts_e2e.py ..                              [100%]

======================= 96 passed in 1379.27s (0:22:59) ========================

mise run test

[test:skill] ........................................................................ [ 85%]
[test:daemon] ........................................................................ [ 26%]
[test:skill] ..................................................                       [100%]
[test:skill] 338 passed in 6.00s
[test:skill] Finished in 6.77s
[test:daemon] ........................................................................ [ 33%]
[test:daemon] ........................................................................ [ 39%]
[test:daemon] ........................................................................ [ 46%]
[test:daemon] ...............................sssssss.................................. [ 53%]
[test:daemon] ........................................................................ [ 59%]
[test:daemon] ........................................................................ [ 66%]
[test:daemon] ........................................................................ [ 73%]
[test:daemon] ........................................................................ [ 79%]
[test:daemon] ........................................................................ [ 86%]
[test:daemon] ........................................................................ [ 92%]
[test:daemon] ........................................................................ [ 99%]
[test:daemon] ....                                                                     [100%]
[test:daemon] 990 passed, 94 skipped in 46.12s
[test:daemon] Finished in 46.98s
Finished in 47.01s

UI evidence

N/A: no UI change (CLI recover output gains an attach: step line).

Checklist

• Test evidence includes the actual commands and output
• UI evidence filled in or marked N/A

🤖 Generated with Claude Code