Claude Skills by Eliyce

Detect Y

Detect Y

TODO

Detect duplicate events across providers

line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line line li...

No description provided

No description provided

No description provided

- `references/missing.md` — nope

- `references/guide.md` — the guide

For more, see references.

For more, see references. // x

Read `references/x.md` before doing Y.

Read `references/x.md` before doing Y. // ok

1. Parse the inventory and normalize each row in prose.

1. Run `scripts/parse-inventory.mjs` to derive the inventory.

Generate acceptance criteria tied to observable outcomes.

Perform rigorous, risk-first review after design and implementation.

When the `analytics_instrumentation` flag is on and the classify-time gate resolved to `instrument`, this skill wires a feature's analytics the tracking-plan-as-code way (issue #279). The LLM only names the event (following the project's one consistent convention) and writes the doc body; the bundled `instrument.mjs` script computes the normalized slug and the module-owned doc path, writes the per-event doc and refreshes the per-module index via the framework's doc-tree primitives, and prints...

Maintains per-module API documentation after endpoint changes

Synchronize implementation changes back into canonical documentation.

Estimate the implementation phase's token footprint from spec, test plan, and affected files; recommend compaction before context starvation.

Pre-implementation scan that classifies how a proposed change affects each consuming module's public surface and recommends coordinated changes.

Authors the small, human-facing Decision Pause packet described by the Decision Pause Contract (in the framework bootstrap AGENT-BOOTSTRAP.md) without the agent hand-authoring the id, timestamps, or JSON. `create` mints a collision-free `D-<ULID>` id and writes the pending packet; `resolve` records the chosen option and rationale and moves it to resolved. A hand-picked sequential `D-{N}` is rejected, so two developers on parallel branches never collide (issue #272). The writer lives in `src/d...

Identify only the canonical docs made stale by the current diff.

Classify each proposed implementation step against the acceptance criteria so scaffolding and over-build are dropped before code is written.

Canonical orchestrator that runs the fast stale-doc detector and dispatches to per-domain doc-sync skills for the affected canonical paths.

Run the post-onboarding documentation creation workflow using the canonical tracker and the current application manifests. Supports two stages — foundation (docs/instructions/** and module-map.yml) and module-docs (docs/modules/** from reviewed map).

Identify failure modes and uncommon but relevant paths.

Maintains per-module error catalogs

Inspect canonical docs and registries before new artifacts are written.

Convert heterogeneous security evidence into stable pentest finding entries.

Maintain shared business terminology in the glossary.

Maintains cross-module integration documentation

Run the deterministic module-attribution extractor over a user prompt and surface any MD-XXXX drafts via the Decision Pause Contract. This is the first half of the Attribution Gate spliced into feature-development.planning (issue #80). Pattern set is finite, framework-owned, and TS-canonical (`src/module-decisions/extractor.ts`); extending it is a framework PR.

Form a hypothesis about which existing module a user prompt belongs to when the deterministic extractor returns nothing. Reads existing module-map.yml features and source paths, scores token overlap, returns a ranked multi-choice draft, and surfaces it via the Decision Pause Contract. Second half of the Attribution Gate (issue #80, Phase 1 §4.3.b). TS-canonical (`src/module-decisions/inferencer.ts`).

Test-driven module-health rollup. Issue #80, Phase 3. Reads the active stack pack's module_health block, parses the declared coverage and test reports (or one provided via --from-report), attributes rows to module slugs via module-map.yml's source globs, computes per-module coverage_pct / tests_passing / tests_failing / tests_total / change_velocity, and writes the extended .paqad/module-health/<slug>.json record with blocked_metrics populated for any signal that could not be computed. The TS...

Whole-project module-health refresh. The agent-facing "update module health" workflow. Runs the test-driven rollup (coverage / tests / change_velocity from the active stack pack's declared reports) and then the session-evidence sync (defect frequency / contract stability / verification status) across every declared module in one pass, writing the extended .paqad/module-health/<slug>.json records. Wraps the existing `paqad-ai module-health rollup` and `paqad-ai module-health sync` verbs — it i...

Retrospective reconciler that scans the source tree against module-map.yml and docs/modules/, emits MM-* findings into .paqad/module-map/drift.json, and surfaces user-approvable deltas. Issue #80, Phase 2. The TS engine lives in `src/module-map/reconciler.ts`; this skill is the agent-side wrapper that invokes it via the CLI and routes accepted deltas through `src/module-decisions/apply.ts` (the only writer of module-map.yml).

Identify performance hazards in a proposed solution or change set before code is written, classified by severity and hot-path placement.

Classify incoming requests into the framework routing model.

Expand incomplete requests into enforceable requirements.

For each story with hard reversibility or wide blast radius, draft an executable rollback procedure before the story leaves planning.

One-shot classifier for rules-as-scripts (issue #89). Reads every rule file under docs/instructions/rules/**, embeds stable RL-<hash> markers, classifies each rule's verifiability (deterministic / heuristic / unverifiable), detects rules already enforced by ESLint/TS/Prettier/existing paqad infra, detects rule conflicts, and writes a draft docs/instructions/rules/rule-script-map.yml for the user to review before `generate rule scripts`. The deterministic mechanics live in src/rule-scripts/; t...

Single entry point for mutating rules in rules-as-scripts (issue #89) — add / edit / remove / downgrade — with a per-rule cascade and no global rebuild. Adding inserts a bullet + mints a stable RL-<hash> id; editing preserves the id and regenerates only that rule's scripts; removing archives the map entry (delayed delete); downgrading flips a rule to unverifiable and drops its scripts. The deterministic file + map mechanics live in src/rule-scripts/{editor,mutate}.ts; this skill orchestrates ...

Generates one or more deterministic .mjs verification scripts (plus pass/fail fixtures) for each rule classified deterministic or heuristic with no existing enforcer in rule-script-map.yml (issue #89). Every script is validated against its own __fixtures__ before registration — a script that misclassifies its fixtures is rejected and surfaced via the Decision Pause Contract, never added to the map. Strict from generation; there is no shadow mode. The deterministic gate lives in src/rule-scrip...

Retrospective drift detector for rules-as-scripts (issue #89). Compares rule-script-map.yml against the rule markdown on disk and the registered scripts, emits RS-* findings into .paqad/scripts/rules/.cache/drift.json, and surfaces user-approvable deltas via the Decision Pause Contract at feature-development planning entry. No LLM judgement — the TS engine is src/rule-scripts/reconciler.ts; this skill is the agent-side wrapper that invokes it and routes resolutions through rule-editor / rule-...

Executes the registered rule scripts diff-scoped during feature-development.checks, aggregates findings into .paqad/scripts/rules/.cache/report.json with hash-cache invalidation, and decides whether the checks stage is blocked (issue #89). No LLM — the TS engine is src/rule-scripts/runner.ts; this skill is the thin agent-side wrapper invoked from the checks stage. deterministic findings block under mode:strict; heuristic findings route to the review stage and never block.

Validate that requested work stays inside the active spec boundary.

Break work into ordered, coherent stories.