ysskrishna/ai-agent-skills

v1.3.0MIT

Curated Agent Skills (SKILL.md): structured thinking lenses, decision support, and brainstorming workflows for AI coding agents.

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

Install support for more coding CLIs - Codex CLI and App, Gemini CLI, Cursor, Antigravity, GitHub Copilot CLI, Factory Droid, Qwen Code, Grok Build, OpenCode, Pi, Kimi Code, Hermes Agent, Devin and Muse, next to the existing Claude Code, skills.sh and gh skill routes. Every route installs straight from this repo with no vendor approval.
ai-agent-skills bundle plugin - all skills in one install, listed in the Claude Code marketplace beside the single-skill plugins. Codex, Droid, Grok, Muse and Cursor read <plugin>/skills/<name>/SKILL.md only, so they need the bundle.
Per-CLI manifests - plugin.json (Agent Plugins 1.0), .agents/plugins/marketplace.json, .codex-plugin/plugin.json, .cursor-plugin/, gemini-extension.json, .kimi-plugin/plugin.json, and package.json with index.js for OpenCode. No skill text changed and nothing runs at session start.
Install guides - README with an expandable section per CLI, and docs/ with one guide per CLI, what was verified and what was not.
Install tests - scripts/test_installs.sh installs the repo into each CLI with a throwaway HOME and checks that every skill is discovered, and test-installs.yml runs it in CI and weekly.
scripts/bump_version.py - sets one version in every manifest.

scripts/validate_repo.py - parses every manifest, requires one version across all of them, and rejects keys that a CLI's schema forbids.
CI - validate-skills.yml also runs on changes to the new manifests and docs/.

Kimi Code cuts skill descriptions to about 250 characters, so a trailing "Skip for ..." clause can be clipped.
Grok and Factory Droid do not load the single-skill plugins. Install the bundle.
Devin, OpenCode 2.x, Cursor's IDE install and the Codex App were not verified. See docs/.

Seven new skills - five-whys, swot-analysis, pre-mortem, tradeoff-analysis, prioritization, fermi-estimation, and thinking-method-selector, a router that picks a method when the user names none and carries a fallback outline for every other skill.
Worked example per skill - skills/<name>/references/example.md for all 17 skills.
Trigger eval queries - evals/<name>.json with 8 should-trigger and 8 near-miss queries per skill, plus evals/README.md.
Repository checks - scripts/validate_repo.py, run by validate-skills.sh: description rules, example and eval files, README / marketplace / plugin keywords / ClawHub slug map in sync, router coverage, equal plugin and marketplace versions.

All descriptions are intent-led - each opens with the method name, then the situations and phrases it fits, then a Skip clause (253-300 characters, was 549-745 with a shared boilerplate prefix that is now banned).
Skill bodies - every skill gains When to use / Skip, a light path for small asks, a gather-first rule, pitfalls, and one tag vocabulary. analytical-thinking and design-thinking no longer contradict their own Setup text. six-thinking-hats defines the opening Blue block, confirms the closing Blue in Custom mode, and replaces the Depth table with the light path. lateral-thinking concept fan follows de Bono (directions, concepts, ideas). strategic-thinking states its time horizons per domain.
README.md - skills grouped by the job to be done with short "use it when" lines, install lines for all 17 skills, usage examples that describe a problem instead of naming a method.
AGENTS.md - new description rules, required example and eval files, self-contained skills, per-owner ClawHub slug policy, calendar skill versions.
CI - validate-skills.yml also runs on changes to evals/, scripts/, README.md and .claude-plugin/.

scripts/publish_clawhub.py - plan reported critical-thinking, design-thinking and lateral-thinking as new because clawhub inspect fails on slugs shared with other owners. State now comes from the registry with the owner parameter, retries on rate limits, stops on unexpected errors, and lists other owners of a slug.
CHANGELOG - [1.1.0] now has its release date.

Thinking-partner skills — Nine new sequential workflows aligned with the six-thinking-hats pattern: critical-thinking, systems-thinking, creative-thinking, strategic-thinking, analytical-thinking, lateral-thinking, design-thinking, first-principles-thinking, and ethical-thinking. Each is registered in README.md and .claude-plugin/marketplace.json.
Skill validation — validate-skills.sh audits each skills/*/SKILL.md against the Agent Skills specification (frontmatter, naming, description length, optional structure). .github/workflows/validate-skills.yml runs it on pushes and pull requests when skills or the script change.

Thinking-partner skills — Removed fixed Modes / Depth tables from the nine skills above; each skill now uses one default pass with optional user-requested tweaks.
six-thinking-hats — Tighter body and refreshed frontmatter description triggers (aligned with the newer skills’ activation phrasing).
README.md — CI, license, and release badges; overview and install guidance (skills.sh first, per-skill npx skills add … --skill …, expanded Claude Code marketplace commands); full skills table; usage examples.
AGENTS.md — Updated authoring and release guidance for the larger catalog.
.claude-plugin/marketplace.json — Catalog entries and descriptions for every skill; plugin keywords extended for discoverability.

Bundled Cursor plugin files — Removed .cursor-plugin/ from the repository (last shipped in v1.0.1 tree); skills remain installable via the Agent Skills standard and other documented paths.
gemini-extension.json — Removed the standalone Gemini extension manifest from the repo root.

Documentation — AGENTS.md now lists every manifest and changelog location to update when cutting a repository release, alongside per-skill metadata.version guidance.
Plugin metadata — Bumped bundled version fields to 1.0.1 in Claude plugin, Cursor plugin, both marketplace manifests, and gemini-extension.json.

six-thinking-hats skill — Six Thinking Hats workflow with hat-set modes (Full, Creative, Risk, Decision, Custom) and depth levels (Quick, Standard, Deep Dive).
Cursor plugin — .cursor-plugin/ metadata and marketplace listing for distributing skills in Cursor.
Claude Code plugin — .claude-plugin/ metadata and marketplace listing for Claude Code.
Repository foundations — README, MIT license, CODEOWNERS, and GitHub Sponsors funding metadata.