Agent Skills: Incident Hotfix Skill

Incident Hotfix Skill

[!IMPORTANT] Mitigate a production incident or urgent regression first, then route to root-cause remediation and a postmortem.

Optional args: slug=<feature>, ticket=<id/url>, mode=interactive|autonomous|channel, channel=<id>, auto_continue=true|false, profile=business|hybrid|technical.

Instructions

When the user asks to perform this workflow, execute the following steps:

Incident Hotfix Workflow

Goal: Stop user-facing harm immediately, then hand off to root-cause discipline instead of debugging live.

Steps

1. Triage:
   • Severity, blast radius, affected environments/markets, and whether the regression is a rollback candidate.
   • Load deploy-release rollback path and prior deployment report if present.
2. Mitigate first:
   • Prefer rollback, feature flag disable, or config revert over a live code fix.
   • Confirm mitigation stopped user-facing harm before investigating root cause.
   • Do not attempt a root-cause code fix under live production pressure.
3. Stabilize:
   • Monitor logs, metrics, errors, and core user flows until signals return to baseline.
   • Record mitigation timestamp, owner, and evidence.
4. Route to remediation:
   • Once stable, hand off to dev-fix for root-cause debugging under the normal Propose -> Approve -> Verify cycle.
   • Do not skip dev-fix's root-cause-before-code requirement just because mitigation already shipped.
5. Route to learning:
   • After verify-work confirms the permanent fix, route to retro-learn for a postmortem.
   • The confirmed root cause becomes a permanent eval case in the skill or workflow that should have prevented it.

Runtime Contract

• Use only for production incidents or urgent regressions with active user-facing harm; non-urgent bugs go directly to dev-fix.
• Required inputs: incident signal (alert, report, or error spike) plus a mitigation path (rollback, flag, or config revert).
• Return BLOCKED only when no mitigation path exists and the incident cannot be safely stopped.

Handoff Payload

• slug, operator_profile (carried, not re-inferred), severity, mitigation applied, stabilization evidence, outcome report, next workflow.

Blocking Questions

• Ask max 3 at a time with a recommended default and 2-3 options.

Output Template

# Incident Hotfix: [Name]

## Severity And Blast Radius

## Mitigation Applied

## Stabilization Evidence

| Signal | Before | After |
| --- | --- | --- |
| [signal] | [value] | [value] |

## Root Cause Handoff

## Outcome Report
{schema_version: 1, run_id: "[run-id]", slug: "[slug]", workflow: incident-hotfix, feature_status: partially_implemented, started_at: "[timestamp]", completed_at: "[timestamp]", requirement_trace: {brd_objectives: [], requirements: [], acceptance_criteria: [], srs: []}, completed_evidence: [], missing_evidence: [], decision_needed: [], recommended_next_workflow: dev-fix, cost: {source: unavailable}, agent: {identity: "[agent-identity]", model: "[model]"}}

## Next Workflow
dev-fix | verify-work | retro-learn

## Cost Report
Call `get_session_cost(workflow="incident-hotfix")` before final handoff.