| name | incident-team |
| description | Dispatch parallel investigation agents during a production incident — timeline builder, root cause analyzer, impact assessor, and mitigation finder. Use for P1/P2 incidents, production down, data integrity issues. |
Incident Team — Parallel Incident Investigation
Overview
Production incidents waste time when investigation is sequential. This skill dispatches four specialized agents simultaneously to cover the full incident picture — timeline, root cause, impact, and mitigation — then synthesizes into an actionable incident brief.
Speed is the priority. Parallel investigation compresses a 40-minute sequential investigation into ~10 minutes.
When to Use
Use for:
- P1 incidents: production down, payment failures, data loss
- P2 incidents: degraded functionality affecting dealers, latency spikes, cascading failures
- Data integrity alerts: duplicate records, missing data, corrupted state
- Security incidents: unauthorized access, credential exposure, unusual traffic patterns
- Any situation where "what broke, when, and who is affected" are all unknown
Do NOT use when:
- Issue is already identified and you just need a fix (go directly to
/debugging-errors)
- It's a non-production environment with no dealer impact
- You already have a clear root cause and just need a rollback
- The issue has been resolved and you're doing a post-mortem
Step 1: Gather Incident Context
Before dispatching, collect:
Incident brief:
- Service(s) affected: [e.g., payment-service, kmanage-api]
- Symptom: [e.g., 500s on POST /payments, RabbitMQ queue depth spiking]
- First reported: [timestamp or "~X minutes ago"]
- Error sample (paste stack trace or log snippet if available):
[paste here]
- Recent deploys you're aware of: [or "unknown"]
- Jira incident ticket (if created): [MYK-XXXX or "not yet"]
If the user provides a Jira incident ticket number, read it with the Atlassian MCP tool before dispatching.
Step 2: Confirm Before Dispatch
Confirm the dispatch plan:
Ready to dispatch the incident team for: [symptom description]
Agents to be dispatched in parallel:
A. Timeline Builder — git log, deploys, Jira tickets → chronological timeline
B. Root Cause Analyzer — error patterns, logs, code → probable root causes
C. Impact Assessor — affected dealers/users, blast radius estimate
D. Mitigation Finder — rollback options, quick fixes, feature flags
Proceed? (yes/no)
Do not dispatch until confirmed.
Step 3: Dispatch Agents in Parallel
Dispatch all four agents simultaneously.
Agent A — Timeline Builder
You are building a chronological incident timeline. Your goal is to establish
WHAT CHANGED in the 48 hours before this incident, in order.
1. Run: git log --since="48 hours ago" --oneline --all -- [affected service paths]
2. Check for recent Jenkins deploys: look for deploy markers in git tags or commit messages
3. Search Jira for recently closed tickets affecting these services:
- Use Atlassian MCP: search for tickets transitioned to Done in the last 48 hours
- Filter by component/service if possible
4. Note any config changes, environment variable updates, or infrastructure changes mentioned
in commit messages or ticket descriptions
Build a timeline in this format:
[timestamp] — [event type: DEPLOY/COMMIT/CONFIG/TICKET] — [description] — [author]
Flag any event that occurred within 30 minutes of the incident start as HIGH SUSPICION.
Context:
- Service(s): [service names]
- Incident start: [timestamp]
- Error: [error description]
Agent B — Root Cause Analyzer
You are analyzing the root cause of a production incident.
1. Read the error/stack trace provided. Identify:
- The specific class and method where the failure originates
- Any chained exceptions (look for "Caused by:" chains)
- Any relevant error codes or messages
2. Search the codebase for the identified class/method:
- Read the surrounding code
- Look for recent changes (git blame or git log -p)
- Check if there are known similar issues in git history
3. Hypothesize up to 3 probable root causes, ranked by likelihood:
- Hypothesis 1: [most likely] — supporting evidence
- Hypothesis 2: [second most likely] — supporting evidence
- Hypothesis 3: [less likely but worth ruling out] — supporting evidence
4. For each hypothesis, state what would CONFIRM or RULE IT OUT.
Context:
- Service(s): [service names]
- Stack trace / error: [error text]
- Recent changes: [provide Timeline Builder output if available, otherwise "unknown"]
Agent C — Impact Assessor
You are assessing the blast radius of a production incident.
1. Identify which dealers/users are affected:
- Is the failure isolated to specific dealer UUIDs, or system-wide?
- Is it affecting all requests or a percentage (check for feature flags, A/B segments)?
- Are there specific workflows impacted (payments, messaging, scheduling)?
2. Estimate scope:
- How many dealers could be affected? (check service's dealer coverage)
- Are any Tier 1 / high-volume dealers in the affected set?
- Is data being silently corrupted (harder to recover) vs. operations failing noisily?
- Are downstream services also failing (cascading impact)?
3. Assess data integrity risk:
- Could this incident have caused duplicate records, missed writes, or inconsistent state?
- If so, which tables/collections and what time range?
4. Rate overall impact:
- CRITICAL: Multiple Tier 1 dealers down, data corruption possible, payments failing
- HIGH: Many dealers affected, core workflow broken but no data loss yet
- MEDIUM: Some dealers affected, degraded experience, no data integrity risk
- LOW: Isolated, cosmetic, or low-traffic path
Context:
- Service(s): [service names]
- Error: [error description]
- Error rate or frequency if known: [e.g., "100% of requests" or "~20 errors/min"]
Agent D — Mitigation Finder
You are finding the fastest path to restore service for a production incident.
1. Rollback option:
- What was the last stable commit/deploy? (use git log)
- Is a rollback via Jenkins feasible right now? What is the rollback risk?
- Are there database migrations that would make rollback unsafe?
2. Quick fix option:
- Based on the error/stack trace, is there a targeted code fix that could be deployed fast?
- Is there a null check, try-catch, or config change that would stop the bleeding?
- Can it be deployed as a hotfix without full CI pipeline?
3. Feature flag / circuit breaker option:
- Is there an existing feature flag that could disable the broken feature?
- Is there a circuit breaker that could be tripped manually?
- Can the affected endpoint be temporarily disabled at the Kong/load balancer level?
4. Mitigation ranking:
- FASTEST: [option + estimated time to implement + risk level]
- SAFEST: [option + estimated time to implement + risk level]
- RECOMMENDED: [your recommendation with reasoning]
Context:
- Service(s): [service names]
- Error: [error description]
- Recent changes: [provide Timeline Builder output if available]
Step 4: Synthesize the Incident Brief
After all agents return, synthesize into a single incident brief. If Agent D found a quick fix, lead with that — responders need the action item first.
Output Format
## Incident Brief
Service: [service name]
Severity: [P1/P2]
Generated: [timestamp]
---
### RECOMMENDED ACTION
[If Agent D has a clear winner, state it here with estimated time]
> Example: "Roll back to commit abc1234 (deployed 2h ago). Estimated: 5 min.
> Risk: LOW — no DB migrations in recent commits."
---
### Timeline
[Agent A output — condensed]
HH:MM — DEPLOY — payment-service v2.4.1 — jenkins-bot (HIGH SUSPICION: 15min before incident)
HH:MM — COMMIT — fix NPE in PaymentProcessor — jane.doe
...
---
### Root Cause Hypothesis
[Agent B output — top 3 ranked]
1. [MOST LIKELY] NullPointerException in PaymentProcessor.process() due to missing null
check introduced in v2.4.1. Evidence: stack trace matches line 142, commit abc1234
changed this method. CONFIRM by reading PaymentProcessor.java:142.
2. [POSSIBLE] ...
---
### Impact
[Agent C output]
Severity: CRITICAL / HIGH / MEDIUM / LOW
Dealers affected: [count or "all" or "unknown"]
Data integrity risk: YES / NO / UNKNOWN
Cascading: [yes/no + which downstream services]
---
### Mitigation Options
[Agent D output]
FASTEST: [option] — [time] — [risk]
SAFEST: [option] — [time] — [risk]
---
### Open Questions
- [anything agents flagged as unknown that needs human input]
Jira-Ready Format
After the incident brief, emit a compact version for pasting into the Jira incident ticket description:
*Incident Brief — [service] — [date]*
*Recommended Action:* [one sentence]
*Timeline (last 48h):*
[bullet list of HIGH SUSPICION events]
*Root Cause (hypothesis):*
[top hypothesis in 2 sentences]
*Impact:*
[severity + dealer count + data integrity risk]
*Mitigation:*
[fastest option + safest option]
Conflict Resolution
When agents return contradictory findings:
- Agent A says no recent deploy, Agent B says code change caused the bug — note both, flag as NEEDS VERIFICATION
- Agent C says impact is LOW, but Agent D says rollback is risky — surface both, let the incident commander decide
- Never silently discard an agent's finding
Key Constraints
- This skill is for active incidents — bias toward speed, not comprehensiveness
- Agents work from what is available; if logs or deploys are inaccessible, they will note gaps
- This skill does NOT execute rollbacks or code fixes — it finds options for humans to decide
- If an agent errors or times out, mark that section as INCOMPLETE and proceed with others