Skip to content

FEATURES.md

Latest commit

 

History

History
160 lines (125 loc) · 8.96 KB

FEATURES.md

File metadata and controls

160 lines (125 loc) · 8.96 KB
Raw

Features

Product-surface reference for the Agent Skills Repository.

Skills Catalog

Catalog snapshot by category. For canonical runtime inventory, use skills.json:

jq '.skills | length' skills.json

GitHub Workflows

Skill Purpose
gh-commit-push-pr Commit, push, and open a Pull Request

Code Review and Quality

Skill Purpose
diagnose Disciplined six-phase debugging loop — build a deterministic pass/fail signal, reproduce, hypothesise, instrument, fix, regression-test
local-review Findings-first local reviews on workspace diffs
error-handling-review Specialist lens for silent failures — empty/over-broad catch blocks, log-and-swallow, unjustified fallbacks, silent retry exhaustion
type-design-review Specialist lens rating a new/changed type's encapsulation and invariants; flags illegal states left representable

Content Creation and Design

Skill Purpose
design-md Read, write, lint, diff, and export DESIGN.md files via the pinned @google/design.md CLI
design-critique Audit UI against a 37-pattern slop catalog and return ranked, scoped findings
frontend-design Distinctive, production-grade frontend interfaces
gemini-imagen Image generation/editing via Gemini API
gpt-imagen Image generation/editing via OpenAI Image API
theme-factory Apply preset or generated theme systems
web-design-guidelines Review UI against web interface guidelines

The four design skills above compose into a pipeline: design-md (token spec) → frontend-design (build) → design-critique (taste / AI-slop review), with web-design-guidelines as a parallel rule-compliance pass (Vercel WIG, accessibility, UX). Each skill's body carries a Sibling skills footer that names the adjacent stages and hand-off cues.

Development Workflows

Skill Purpose
api-design Design and review robust API, event, interface, and machine-output contracts
loop-design Design verifiable autonomous loops on top of /loop and /goal; gate on an oracle, then scaffold a portable loop bundle (/loop-design)
brainstorming Clarify WHAT to build and capture the chosen direction (docs/design/)
write-spec Pin the falsifiable contract — WHAT must be true, mechanism-free; conditionally add high-risk authority/invariant scenarios and review closure (docs/specs/)
write-plan Sequence the build — HOW: tasks, files, seam selection, and verification; conditionally add high-risk traceability, evidence, capability gates, and review closure (docs/plans/)
blind-spots Find what you don't know about one change — blind spot pass before it's built, or a brief-then-quiz before you merge (never scored, never gating)
create-cli CLI parameter and UX design
agent-native-architecture Build agent-native apps with tool/action parity
deep-research Web-backed research with conditional depth, explainable URL-host evidence filtering, and verified first-party source safeguards
research-architect Engineer deep-research prompts, route execution, independently verify reports, synthesize multi-run results, and compound lessons via postmortems

Disciplines

reference-typed skills that govern how the agent operates rather than what to build. They produce no discrete deliverable; they shape style, rigor, and reasoning across every other skill.

Skill Purpose
caveman Sticky ultra-compressed output mode (~75% token cut) until the user says "stop caveman"
first-principles Systems reasoning for high-stakes decisions — epistemic framework, decision matrix, principle tensions
test-strategy Testing methodology: TDD, real dependencies over mocks, behavior-based tests, and conditional effective authority-boundary probes
verify-before-complete Evidence-based completion gate; require verification before claiming done

Security

Skill Purpose
repo-hardening Audit repos for supply-chain risk, generate repo-local security artifacts, and drive hardening work
secure-code Static analysis security scanning and architectural trifecta detection via semgrep
audit-skill Security-audit agent skills for prompt injection, overreach, secrets, and dangerous code patterns

Platform Integrations

Skill Purpose
vercel-deploy Deploy to Vercel
vercel-preview-logs Debug Vercel preview deployments
vercel-react-best-practices React/Next.js performance optimization
nextjs-app-router Next.js App Router patterns, data fetching, and debugging
vercel-composition-patterns Scalable React composition patterns

UI Automation

Skill Purpose
playwright Browser automation from the terminal — navigation, form filling, screenshots, data extraction
screenshot Capture screenshots across platforms (full screen, app/window, region)

Knowledge and Documentation

Skill Purpose
obsidian-markdown Obsidian Flavored Markdown with wikilinks and callouts
obsidian-bases Obsidian Bases with views, filters, and formulas
obsidian-canvas Obsidian Canvas files for visual canvases
markdown-converter Convert file formats to Markdown
fetchmd Convert webpages or local HTML into clean Markdown for AI workflows
handoff Session summaries for context handoff
compound-docs Capture solved problems as documentation
session-retro Route non-obvious session learnings into existing project reference docs

Meta / Skill Management

Skill Purpose
skill-creator Initialize, validate, and package new skills
skill-evals Validate SKILL contract compliance and run trigger-eval scaffolds
skill-installer Install skills from curated lists or GitHub repos
skill-standardizer Detect drift and unify skill copies, with optional single-skill targeting
find-skills Discover installable skills
template Starter scaffold for new skills

Hook-Enforced Quality Gates

  • SKILL.md frontmatter validated on every write/edit (pre-tool-use).
  • New or updated skills should declare skill-type so contract validation applies the right structure for workflow vs reference skills.
  • Every skill declares a SemVer version; release-relevant skill edits require a version bump and changelog entry against the selected git base.
  • Pushes to protected branches are blocked unless the command includes an explicit DOJO_ALLOW_PROTECTED_PUSH=1 override.
  • skills.json manifest regenerated after every SKILL.md change (post-tool-use).
  • Newly authored design summaries, specs, and plans identify the producing agent in author: frontmatter; current-schema spec/plan validation rejects missing or unresolved attribution while accepting legacy artifacts.
  • docs/specs/*-spec.md and docs/plans/*-plan.md validated after every write/edit (post-tool-use).
  • Uncommitted changes and untracked files blocked at session stop. Unpushed commits no longer block — push timing is the operator's call.
  • Skill directory structure and skill release-version bumps validated at session stop.

Command Wrappers

Slash-style entrypoints for harnesses that support command files. scripts/gen_harness_adapters.py links each skill's commands/<rel>.md into .claude/commands/<rel>.md (local-only, gitignored), so Claude Code resolves them as real slash commands — nested files like commands/workflows/brainstorm.md become /workflows:brainstorm. Run the generator after a clone to populate them:

  • /review — local code review
  • /review-pr — GitHub PR review
  • /fix-issue — GitHub issue resolution
  • /triage-issue — GitHub issue triage
  • /commit-push-pr — commit, push, PR flow
  • /workflows:brainstorm — brainstorming session (WHAT — chosen direction)
  • /workflows:spec — write the falsifiable contract (WHAT must be true)
  • /workflows:plan — sequence the build (HOW — task breakdown, files, steps)
  • /understand-change — blind spot pass on a proposed change: scope, blast radius, and unknown unknowns
  • /quiz-change — get briefed on an implemented change, then quizzed on it one question at a time
  • /standardize-skills — skill standardization
  • /deep-research — route depth and filter evidence in one command
  • /repo-audit — generate repo-local audit artifacts and summarize hardening gaps
  • /repo-harden — implement the highest-value hardening fixes and refresh the audit packet
  • /retro — capture session learnings into existing project reference docs, including matching root, docs/system, or docs/project files

Skill Packaging

Skills can be packaged into .skill files (zip format) for distribution:

python skills/skill-creator/scripts/package_skill.py <skill-name> ./dist

Pre-packaged skills are available in docs/downloads/.