emergency-fix

name	emergency-fix
description	FORCE-MODE hotfix: smallest patch → commit → push DIRECTLY to main → follow-up ticket. Needs TERMINAL_FORCE_MAIN=1 (hook-enforced). Use on /emergency-fix or 'this is breaking prod' / 'hotfix' / 'we need to revert now'.

/emergency-fix — production hotfix bypass

/emergency-fix is the one and only skill that intentionally bypasses the human-merge gate on main/master. It is for production-impacting bugs that are hurting real users RIGHT NOW. Every other change goes through /pr-creation → code-review agent → human merge.

[1] Authority

The skill ONLY works when the caller is launched with TERMINAL_FORCE_MAIN=1 in the environment. Without it, the global ~/.claude/hooks/block-main-merge.sh PreToolUse hook will refuse the final git push origin main. Three legitimate ways to be authorized:

1. TerMinal's emergency-fix / unblock-ci / revert-main agents — the runner injects the var per-spawn.
2. A deliberate manual launch: TERMINAL_FORCE_MAIN=1 claude from the shell, with a written reason in your scrollback. Do NOT export the var in ~/.zshrc — that's how accidents happen.
3. autopilot-harness — the harness's own repo is already exempt from the hook by path; the force-mode mechanism is irrelevant there.

If you are NOT in one of those three contexts, this skill is the wrong tool — use /pr-creation instead.

[2] Sequence

   1. Reproduce the failure (briefly — confirm scope before patching)
        │
        ▼
   2. Write the SMALLEST surgical change that stops the bleeding
        │  · no refactor · no cleanup · no scope expansion
        ▼
   3. Run the relevant test subset; confirm green
        │
        ▼
   4. Commit on main with `fix: <subject>` (Conventional Commits)
        │
        ▼
   5. `git push origin main`     ← hook bypassed via TERMINAL_FORCE_MAIN=1
        │
        ▼
   6. File a backlog ticket capturing real root-cause + proper fix prompt
        │  · type: bug · priority: high · source: emergency-fix
        ▼
   7. Hand off with the SHA you pushed + ticket id

[3] Bar — when NOT to use

• The bug is not actually breaking production. (→ /pr-creation)
• The minimal patch isn't obvious within ~5 minutes of investigation. Page humans instead — a wrong hotfix is more expensive than a paged on-call. File a critical ticket and stop.
• The fix touches more than one logical change. Split it.
• You're tempted to "also clean up X while I'm here." No. Surgical.

[4] What gets committed

• The minimal source change (one file is the norm; two is the cap).
• A test that pins the regression, IF the smallest-possible-test fits into the same patch. If a real test takes a refactor to write, ship the fix now and file the test as the follow-up ticket.

[5] What goes in the follow-up ticket

• Title: "Proper fix: "
• type: bug / priority: high / source: emergency-fix
• owner: exactly one assigned agent (agent_id, agent_scope, agent_kind) selected with list_agents
• Body must include:
  • The SHA pushed to main as the emergency patch
  • Why the patch is a stopgap (what it papered over vs what's still broken)
  • A self-contained agent-runnable fix prompt for /factory to take
  • Any test coverage gap the emergency revealed

[6] Related FORCE-mode skills (also bypass main)

• /unblock-ci — main CI is red blocking the team; narrow revert/pin/skip.
• /revert-main — narrowly git revert HEAD on main and push.

All three flow through the same TERMINAL_FORCE_MAIN=1 carve-out. The authority is the same; the playbook differs.