From cadence
Use when an issue needs deeper analysis than triage — confirming root cause by reading code, verifying file/line references in body are still valid, checking blast radius across the codebase, or proposing a concrete fix direction. One issue per invocation.
How this skill is triggered — by the user, by Claude, or both
Slash command
/cadence:investigate-issueThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Investigation goes beyond labels to verify the issue's claims against actual code, confirm root cause, and propose a fix direction. Output is a comment that either:
Investigation goes beyond labels to verify the issue's claims against actual code, confirm root cause, and propose a fix direction. Output is a comment that either:
status/needs-info and stopInvestigation ≠ triage: 30 min – hours per issue, not 5 min. Don't run it on the whole backlog. Pick issues that earn it.
Don't use when light triage is enough (body is thorough, claims unambiguous, only need labels).
Read the body completely — note every claim:
Verify the claims — open each file:line:
Probe blast radius — don't trust the body's enumeration:
grep / Glob for the same anti-pattern elsewheregit blame on suspect lines — recent change? old?Decide outcome:
status/needs-infoApply outcome:
assignees, link to draft PRMatch repo's PR template style — emoji-headed sections for at-a-glance scanning. Match issue body's language. Skip sections that don't apply (don't leave empty headers).
## 🔬 Investigation YYYY-MM-DD
**Verdict**: confirmed · dismissed · needs-info
**Adjusted priority**: <only if triage call changed after investigation>
### 📊 At a Glance
<diagram (mermaid or ascii) + 1-line caption — show the verified mechanism, not a text summary>
Use a diagram when investigating:
- Security: `sequenceDiagram` of attack chain confirming where the actual exploit happens
- Race condition: `sequenceDiagram` of concurrent timeline
- Refactor: `flowchart` of current vs proposed module shape
- Cluster of broken-together bugs: `flowchart LR` of dependency
Skip the section if the verdict line + Verified bullets already make the picture obvious.
### ✅ Verified
- <file:line still valid as of `<sha>` / matches body's description>
- <reproduction confirmed via `<command / scenario>`>
- <blast radius: N additional sites at `<paths>`>
### ❌ Body inaccuracies
<only if applicable — e.g. "body says fix is at A, actual root cause is B because...">
### 🛠 Proposed Direction
<2-4 lines — concrete fix path. File paths + shape of change. Not full code.>
### ❓ Open Questions
<only when investigation surfaces decisions / tradeoffs / verification gaps that the issue owner needs to resolve before fix can land>
Format each as: `**<question>** — <type>. <your lean if any>`
Types: `decision needed` · `tradeoff` · `verification gap`
Skip the section when there's nothing genuinely undecided. Forced "questions" become noise.
### 🔗 Related
<cross-links if relevant>
| Triage | Investigation | |
|---|---|---|
| Reads body | yes | yes |
| Reads code | no | yes |
| Verifies file:line | no | yes |
| Probes blast radius | no | yes |
| Proposes fix | only echoes body | own analysis |
| Time / issue | 2–10 min | 30 min – hours |
| Scope | whole backlog | one issue at a time |
If you start reading code during a triage, you've crossed into investigation — finish that one issue properly. Don't half-investigate.
status/needs-info and stop.npx claudepluginhub kirkchen/cadence --plugin cadenceGuides creation, editing, and verification of skills for AI coding agents using test-driven development with subagent scenarios. Use when authoring or debugging skills.