[rollback] | Structured Rollback

Purpose

Answer: "How do we safely reverse a bad release and verify recovery?"

When to Use

• Production incident requiring immediate revert
• Failed deployment or broken CI on main
• After [incident] identifies a bad commit
• Health probes reporting failures after a merge
• Release promotion must be reversed before broader rollout

Workflow

0. Emit SKILL_INVOKE

Post SKILL_INVOKE to the bus before any stateful action.

Type: SKILL_INVOKE
To: all
Message: [skill=rollback] [mode=side_effecting] [args_hash=<sha256>] [session=<session_id>]

(The skill invocation runtime injects the caller's canonical identity as from_id.)

1. Identify scope -- parse $ARGUMENTS for commit SHA, PR number, or "last N commits"
2. Find last known good -- check CI status, health probes, and test results

   git log --oneline -10
   python -m hummbl_governance.services.health --check 2>/dev/null || echo "HEALTH: DEGRADED"
   

3. Dry run (if --dry-run or by default on >3 commits):
   • Show files affected: git diff --stat <good>..<bad>
   • List services touched
   • Flag contract schema changes (requires SemVer bump)
4. Execute revert:

   git revert --no-edit <bad-commit>..HEAD  # or single commit
   

5. Verify health:

   python -m pytest $PROJECT_ROOT/tests/test_acceptance.py -v
   python -m hummbl_governance.services.health --check
   

6. Notify bus: Post STATUS to the bus with the rollback result.

   Type: STATUS
   To: all
   Message: ROLLBACK: reverted <sha> -- health restored
   

   (The skill invocation runtime injects the caller's canonical identity as from_id.)
7. If contract baseline affected, warn: contract rollbacks may require downstream coordination.

Output Format

[rollback] | <commit-or-range>
------------------------------
Target:       <commit SHA or range>
Reason:       <from incident or argument>
Files:        <N files changed>
Services:     <list of affected services>
Contracts:    <SAFE | AFFECTED (requires coordination)>
Health before: <PASS | FAIL (N probes)>
Revert:       <commit SHA of revert>
Health after:  <PASS | FAIL (N probes)>
Tests:        <PASS | N failures>
Bus:          <posted | skipped>
------------------------------
Status: ROLLED BACK | DRY RUN ONLY | FAILED (reason)

Skill Chains

Mandatory (MUST pass before rollback execution)

• [deploy-health] MUST confirm degradation — no rollback without evidence of failure Exception: operator explicit override for emergency rollback

Advisory

• [incident] -> [rollback] (incident identifies, rollback executes)
• [rollback] -> [health] (verify recovery)
• [rollback] -> [sitrep] (communicate status to team)
• [rollback] -> [retrospective] (after stabilization)
• [release-manager] -> [rollback] (release plan includes recovery path)

Authority

• T1 (TRUSTED): May rollback with [deploy-health] confirming degradation
• T2 (Active/High): MUST get operator approval AND [deploy-health] confirming degradation
• T3 (Medium): MUST get operator approval AND [deploy-health] confirming degradation
• T4 (Probationary): BLOCKED — cannot invoke this skill
• Operator: Override any restriction — emergency rollback allowed without pre-chain