Menu
Use when you need deep analysis of code, bugs, performance, or architecture issues.
Installer avec Codex ou Claude Copiez ce prompt, collez-le dans Codex, Claude ou un autre assistant, puis laissez-le vérifier la page du skill et l'installer pour vous.
Basé sur la classification professionnelle SOC
Use when you have a large-scale, multi-phase project requiring orchestrated execution.
Use when you need persistent looping until a task is fully complete.
Deterministic adversarial code review for high-stakes scope — independent per-dimension review, a non-skippable per-finding refutation, completeness + interaction critics, and a deterministic 3-lens rubric judge panel. Opt-in main-loop Workflow tier.
Use when the user wants to review accumulated skill feedback, verdict trends, or improvement candidates collected during Stage 4 retrospectives. Trigger on "show skill feedback", "스킬 피드백 보여줘", or finding which skills need /writing-skills work.
Deterministic competitive code tournament — N builders independently solve one task and return patches, an Action-Kamen judge scores them head-to-head, the winner is picked by score and applied. Opt-in main-loop Workflow tier.
Deterministic adversarial debate for high-stakes or irreversible decisions — mandatory refutation plus a scored judge panel. Opt-in main-loop Workflow tier.
| name | team-shinchan:analyze |
| description | Use when you need deep analysis of code, bugs, performance, or architecture issues. |
| user-invocable | false |
If args is empty or only whitespace:
Ask user: "What would you like to analyze?"
STOP and wait for user response
If args length > 2000 characters:
Truncate to 2000 characters
Warn user: "Request was truncated to 2000 characters"
If args contains the flag --single, skip the panel and run the Fallback single-Hiroshi Task at the bottom. Otherwise run the diverse-lens panel. (Match the literal flag only — do NOT substring-match words like "quick", which legitimately appear in targets such as "analyze the quicksort path".)
Quality-first: a single analyzer sees one frame and systematically misses what that frame can't see (a structural pass overlooks the auth edge case; a correctness pass overlooks the architectural smell). Launch independent read-only lenses in PARALLEL, then synthesize.
// Lens 1 — architecture / root-cause
Task(subagent_type="team-shinchan:hiroshi", model="opus",
prompt=`/team-shinchan:analyze (architecture & root-cause lens). READ-ONLY: do not modify any file.
Analyze: ${args}
Focus: structure, dependencies, design seams; if a bug, the most likely root cause with evidence.
Return concrete, file:line-anchored findings + recommended direction.`)
// Lens 2 — correctness / security / failure-mode
Task(subagent_type="team-shinchan:actionkamen", model="opus",
prompt=`/team-shinchan:analyze (correctness, security & failure-mode lens). READ-ONLY: do not modify any file.
Analyze: ${args}
Focus: logic errors, edge cases, injection/authz, data-loss and failure paths.
Return concrete, file:line-anchored findings.`)
// Lens 3 — hidden requirements / edge cases / implicit assumptions
Task(subagent_type="team-shinchan:misae", model="opus",
prompt=`/team-shinchan:analyze (hidden-requirements & edge-case lens). READ-ONLY CONTRACT: analysis only — do NOT write any file, do NOT touch WORKFLOW_STATE.yaml / REQUESTS.md, do NOT treat this as Stage 1, do NOT spawn Tasks.
Analyze: ${args}
Focus: unstated requirements, implicit assumptions, edge/empty/error states the happy path ignores.
Return concrete findings.`)
If the target is a bug or incident, ALSO launch a code-flow trace in parallel:
Task(subagent_type="team-shinchan:shiro", model="sonnet",
prompt=`/team-shinchan:analyze (code-flow trace). READ-ONLY.
Trace the call path relevant to: ${args}. Return the execution path file:line by file:line.`)
One Hiroshi pass reconciles the lenses (resolve disagreements — do not just concatenate):
Task(subagent_type="team-shinchan:hiroshi", model="opus",
prompt=`/team-shinchan:analyze synthesis. Reconcile these independent lens analyses into ONE report and explicitly resolve any DISAGREEMENTS between lenses.
Lens findings:
${lens_outputs}
Produce: | Current state | Issues discovered (ranked) | Recommended solutions | File/line references |.`)
--single / delegated)Task(
subagent_type="team-shinchan:hiroshi",
model="opus",
prompt=`/team-shinchan:analyze has been invoked.
## Deep Analysis Request
Perform the following types of analysis:
| Type | Analysis Content |
|------|----------|
| Code Analysis | Structure, dependencies, complexity |
| Bug Analysis | Error cause, stack trace, reproduction conditions |
| Performance Analysis | Bottlenecks, memory, optimization strategies |
| Architecture Analysis | Overall structure, improvements, tradeoffs |
## Result Requirements
- Current state summary
- Issues discovered
- Recommended solutions
- Related file and line references
User request: ${args || '(Please describe what to analyze)'}
`
)