notesage

Fill missing type, title, and description frontmatter on documents using structured AI output

Download a web page by URL and save it as clean markdown with images

Audit for accessibility — keyboard navigation, ARIA labels, contrast, focus indicators

Produce a full dependency health report (SBOM, vulnerabilities, staleness, upgrades, licenses)

Audit error handling UX — error boundaries, silent failures, loading states, empty states

Audit for unnecessary re-renders — Zustand subscriptions, missing memoization, inline callbacks

Audit test coverage — inventory tests by type, find critical untested paths

CI-repair stage of the AW pipeline. Diagnoses a red CI on a bot-authored draft PR and applies a single bounded, well-understood mechanical fix (one attempt only); posts a comment for any failure pattern it cannot safely repair.

Push a small follow-up commit on an existing draft PR branch in response to specific human feedback. Same hard gates as aw-tdd. Deflects to label-reset if the requested change is too large or scope-changing.

After a claude-authored PR is merged, look for divergence between what the originating skill prescribed and what actually shipped. Propose skill patches as a draft docs PR. Never auto-merges — human reviews every proposed change.

Implement a `tdd + afk` issue using TDD red-green-refactor. The issue is either an issue that aw-slice did not split (the common case) OR one of the peer issues created by aw-slice when an issue had multiple independent user values. Writes failing tests first, implements minimum to pass, runs full test suite as a hard gate, opens a draft PR. Updates labels through tdd → review. Reverts and reports if anything fails.

Prepare a release with version bump, changelog, and history entry

Use when adding new Tauri IPC commands (Rust backend functions callable from the frontend), modifying existing commands, or working with Tauri plugins.

Use when modifying markdown parsing, serialization, Tiptap schema changes, or any code that affects how markdown is converted to/from the ProseMirror document model.

Run performance benchmarks and update the baseline

Dispatch to the right test skill based on scope or what changed

Use when creating or modifying Tiptap editor extensions, ProseMirror plugins, or working with the editor's document model, decorations, or transactions.

Use when creating, scaffolding, modifying, or styling any UI component. Ensures shadcn/ui is used first, design system compliance, proper theming, and polished visual quality.

Verify implementation against PRD requirements and quality gates

Audit for race conditions, stale closures, missing cancellation, and error handling gaps

Audit for unused exports, dead code, unused dependencies, and deprecated code

Audit for documentation drift — stale file paths, wrong signatures, outdated descriptions

Audit for oversized components, hooks, and modules that need decomposition

Audit for memory leaks — event listeners, intervals, process cleanup, stale refs

Audit Rust backend for mutex issues, panics, process management, and concurrency

Audit for security vulnerabilities — injection, XSS, credentials, sandboxing

Audit for type system abuse — any types, unsafe casts, missing types, ts-ignore

When a human comments on a hitl-labeled issue OR comments/reviews a claude-authored PR, interpret the feedback and take the appropriate label/state action. Does not generate code — only translates feedback into pipeline state changes.

Rewrite a refine-flagged GitHub issue body into the outcome-oriented template (bug / enhancement / chore variants), preserving all reproduction details. Adds `refined` (state) + `slice` (action), removes `refine`. Runs after `aw-triage` has classified the issue.

Independent reviewer of a draft PR opened by aw-tdd. Reads the issue body PLUS every comment posted after the latest `refined` marker, reads the PR diff and tests, and verifies that EVERY acceptance criterion AND every user-comment scope-change is satisfied. Posts a per-criterion checklist; if gaps are found, closes the PR and resets the issue to tdd+afk so aw-tdd can retry with the gap list as context.

Decide whether a refined GitHub issue ships as one PR or as N PRs. Default is one PR (no slicing) — issues with a single user value pass directly to aw-tdd. Slice only when the issue delivers N independent user values, each independently shippable. Creates a research peer issue first if under-specified.

Classify a fresh GitHub issue (bug / enhancement / chore / duplicate / wontfix), set the category label, and add the `refine` action label so aw-refine picks it up next.

Save a behavioural correction the operator just gave into `.claude/feedback/` as a new rule. Writes the file with proper frontmatter, regenerates `.claude/feedback/INDEX.md` and the per-skill curated sections, and stages the changes for commit. Use when the operator says "save this as feedback", "remember this rule for AW", or similar.

Generate PowerPoint presentations from documents

Save and organize research files with metadata and tags

Search saved research files by tag, keyword, or content

Synthesize findings across multiple research sources

Implement tasks from a task breakdown file with parallel sub-agents, test gates, and doc updates

Break a PRD or feature into numbered implementation tasks

Create a Product Requirements Document for a new feature