[aarontrowbridge]

Related Issue

None filed yet — this is a draft from the skills-integrity loop. Design-of-record: proposal S-fcz-02 + findings F-fcz-07/08/09/15 (personal vault amicode/skills-integrity/findings/). File an issue before marking ready (PR template requires one).

Description

New public skill detached-run: a bounded, stoppable background launcher (setsid + nohup + timeout -k + nice, PID file recorded from inside the new session, MemAvailable gate, kill-by-recorded-PID process group), with the traps that five hand-copied launchers in one campaign re-derived: free vs available column, timeout without -k, killing only the wrapper PID, never pkill -f on a shared box, block-buffered stdout, load-dependent timings.

Evidence

Fluxonium-cz campaign observation notes B1, B6, B7, B8, H2c, D2a, G1, B11 (fluxonium-cz-intermediate/ledger/observations/), friction note F-fcz-07/08/09/15.

Honesty flags (also in the card)

• Draft, partially verified: launcher and stop scripts were executed with a SIGTERM-ignoring stand-in child; the card lists what was and was not checked (orphan-on-wrapper-kill reproduced; timeout without -k leaves a SIGTERM-ignoring child; -k 60 reaps; memory gate and re-launch refusal; survives caller exit).
• Unverified: why the real Julia child outlived timeout's SIGTERM in the campaign (mechanism not established); macOS behaviour (Linux /proc assumed); the amico run quirks (Julia-only, env stripping, orphaned child) are the campaign's observations, not a current spec.
• References the internal live-solve skill only conditionally ("if your session has it").

Verification

cd packages/extension && node scripts/lint-skills.mjs --known scripts/lint-skills-known.txt → lint-skills: 41 skills, 0 error(s), 5 warning(s) — ok (the 5 warnings are pre-existing, none in this card). Also run with the unmerged private-package rule from origin/private-skills-out (--denylist): 0 hits in this card. Tests: skills_integrity, skill_salvage_851, workflow_skills_public 42/42 pass; the full vitest run shows the same 128 failures as clean origin/main in this environment (fleet/sidebar/pack tests; unrelated) plus one flaky fake_serve_reaper case that passes alone.

[coderabbitai]

Important

Draft PR not reviewed

Draft PRs are not automatically reviewed by default.

• Trigger a manual review

To automatically review draft PRs, update your CodeRabbit configuration:

reviews:
  auto_review:
    drafts: true

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

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