analyze capability

---

name: analyze description: Use when assessing a codebase or diff for quality, security, performance, or architecture problems before review or shipping — surfaces decidable findings as graph artefacts. allowed-tools:

• mcp__plugin_agency_agency__search
• mcp__plugin_agency_agency__get_schema
• mcp__plugin_agency_agency__execute
• Bash

---

analyze capability

Analyze runs decidable transforms over source and reports findings on the quality, security, performance, and architecture axes as graph nodes the orchestrator can reason about, rather than prose opinions.

When to use

• Unsure whether a change is safe to ship
• Suspected security or performance regressions in a diff
• A codebase area that feels risky or unfamiliar before review

Verbs

Verb	Role	Brief	Reference
architecture	transform	Dependency-graph + structural checks: import cycles, file LOC thresholds.	details
cleanup	act	Focused mode: analyse for dead-code findings only, draft a patch plan.	details
graph	transform	Query the provenance graph — a census of node types + a typed listing (read the graph).	details
improve	act	Read prior Analysis findings, draft an improvement plan as a Reflection.	details
paths	transform	Spec 048 intent-path analysis: long chains + verb sequences.	details
performance	transform	AST-based hot-path lint: nested O(n²), += in loop, unbounded while True.	details
quality	transform	Decidable lint findings: unused imports, long lines, long functions, long files.	details
run	act	Run the requested axes; record an Analysis + per-Finding nodes.	details
security	transform	Decidable security patterns: eval/exec, hardcoded credentials, pickle.load, shell=True.	details

Example

await call_tool('capability_analyze_architecture', {'intent_id': 'intent:abc'})

Red flags — stop and re-read this skill

• Shipping a risky diff with no analysis → run capability_analyze_security first
• Hand-waving 'looks fine' on unfamiliar code → get findings via capability_analyze_quality

Walk this capability

Drive this capability's verbs by WALKING a skill one phase at a time (progressive disclosure, recorded as provenance):

• code-analysis (discipline): scope → axes → run → review → apply — walk it: await call_tool('capability_develop_skill_walk', {'name': 'code-analysis', 'inputs': {}, 'intent_id': '…'})