Use when auditing a TypeScript/JavaScript codebase for dead code and unused dependencies — unreferenced files, dead exports, orphaned modules, unused npm packages, phantom deps — and want a report separating safe removals from ones needing confirmation, then chunked cleanup. Keywords knip, depcheck, ts-prune, unused imports, unreachable code.
Shared investigative substrate for evidence-grounded, read-only analysis skills — provides capability-role source discovery, the evidence-vs-hypothesis standard, the parallel-subagent findings-digest schema, adversarial verification, the app-version/release dimension, local-markdown artifact conventions, and the optional ticketing flow. Invoked BY analytics-friction-analysis, error-triage, regression-analysis, and evidence-consolidation; NOT used directly and NOT a target for direct user requests.
Use when merging the findings of two or more evidence-analysis reports (analytics-friction-analysis, error-triage, regression-analysis) into one reconciled report — deduping overlapping findings, letting each domain's evidence raise or lower the other's confidence, and producing a single ranking plus a merged "considered & ruled out". Defaults to reading existing report artifacts (cheap); can optionally run the leaf skills fresh first. NOT for a single-domain analysis — invoke the relevant leaf directly.
Use when starting a new feature from a ticket or spec and need a technical design document before implementation. Outputs an authoritative design doc that locks in scope, alternatives considered, file layout, edge cases, and out-of-scope before any plan or code is written.
Use after a feature design doc is approved and an implementation plan is needed. Outputs a phased plan with explicit MANUAL REVIEW CHECKPOINT gates between phases (one commit per phase, not per task) and a paired risk-ordered QA checklist.
Use when reviewing a GitHub pull request or a diff with multiple specialized review agents and you want every finding verified against the actual code before it is reported. Triggers on "review this PR", "kick off a PR review", pre-merge review requests.
Use when investigating a metric-grounded regression — a measured worsening of latency, error rate, throughput, or similar tied to telemetry, a specific metric/dashboard, or an alert/incident. Fans out across the observability MCP servers the user has configured to produce evidence-ranked root-cause hypotheses. NOT for proactively mining product analytics funnels (use analytics-friction-analysis), NOT for digesting an error tracker (use error-triage), and NOT for speculative "feels slow" perf concerns during feature development — route those to systematic-debugging / investigate. To merge this report with another domain's, use evidence-consolidation.
Use when upgrading, updating, or bumping npm dependencies, especially major version upgrades spanning multiple packages or sessions. Also use when user asks to update outdated packages, resolve peer dependency conflicts, or plan a dependency migration.