A third-party marketplace of language-agnostic AI workflow plugins. Plugins
install into Claude Code
and Codex, and are
readable by agents scanning .agents/skills/.
Warning: Review plugins before installing. Anthropic does not control plugin contents and cannot verify they work as intended.
Repository: tony/skills
| Plugin | Category | Description |
|---|---|---|
| commit | Development | Create git commits following project conventions with format enforcement and safety checks. |
| weave | Development | Run independent, adversarial AI participants via sub-agents or model CLIs to brainstorm, refine, plan, execute, architect, review, and synthesize. |
| rebase | Development | Automated rebase onto trunk with conflict prediction, resolution, and quality gate verification. |
| changelog | Productivity | Generate and maintain categorized changelog entries from branch commits and PR context. |
| tdd | Testing | TDD bug-fix workflow: reproduce bugs as failing tests, determine root cause, fix, and verify. |
| model-cli | Development | Run prompts through individual AI CLIs — Antigravity/agy (Gemini), codex/GPT, and cursor/agent with fallback support. |
| pr | Development | Manage gold-standard PR descriptions. Detects AI slop and verbose commits, resolving them via fixup commits and autosquash. |
| research | Learning | Study dependencies locally. Clones upstream repos and creates version-pinned worktrees matching your project's exact versions. |
| slop | Development | Scan tracked files for AI slop and verbose noise, resolving each finding with atomic, verified commits. |
| tailwind | Design | Detect and fix inconsistent spacing, margin/gap mixing, and padding asymmetry in Tailwind CSS layouts. |
| pytest-optimizer | Testing | Profile and optimize pytest suites. Ranks and applies safe speedups as verified, independent commits. |
| spike | Development | Run no-commit spikes, strategy bakeoffs, or multi-round convergence in git worktrees. Tests ideas against quality gates and returns a commit-by-commit implementation plan. |
| respond | Development | Screen review feedback from humans or bots before changing code. Verifies claims against facts and project decisions. Valid findings become gated, atomic commits; invalid ones get evidence-backed replies. |
| action | Development | Convert tickets to branches in isolated worktrees. Uses team branch conventions, lands gated commits, and supports parallel tickets. |
| lean | Development | Writing discipline and cleanup tools for tight, slop-free prose and code. |
| double-check | Productivity | Forces verification requests to return re-derived answers instead of diffs against prior turns. Includes an alignment tool for repairing chats. |
| release | Development | Cut and bump releases with safe defaults (no automatic push/tag). Rolls out releases to downstream consumers with CI verification. |
| merge-pr | Development | Merge PRs matching repo history conventions. Includes readiness checks, CI watching, stack detection, and automated rebasing. |
| business | Productivity | Measure and report the business value of AI workflows with provenance-tagged data — metrics in engineer-hours and cycle time, never currency. |
| disk | Productivity | Safely reclaim disk space. Classifies consumers as cache, redundant copy, or history, deleting only what is proven safe. |
| github-actions | Development | Update GitHub Actions pins fleet-wide. Verifies tags, researches release notes, commits each action separately, and supersedes Dependabot PRs. |
| situate | Development | Gain situational awareness before modifying code. Scans branches, PRs, tickets, and project conventions to orient the agent and verify the work required. |
| terraform | Development | Upgrade Terraform and OpenTofu versions. Discovers every root module, moves provider constraints together because they combine across modules, and refreshes lock files. |
| ruff | Development | Upgrade Ruff across repositories. Works out which new rules can fire against each repo's select list and lands one reviewed commit per rule, citing the upstream rule doc. |
| git-branch | Development | Rebuild branch history into atomic commits (byte-identical), or reimplement from scratch using existing tests as the spec. Includes an interactive-rebase toolkit. |
| package-updater | Development | Update dependencies and toolchains across repositories. Checks the supply-chain cooldown first, then commits toolchain, named bumps, and lockfile refreshes separately with release notes cited. |
| gh | Development | File durable GitHub issues with reproductions, pinned links, and stripped PII. Enforces a markdown writing discipline reusable for any PR, comment, or ticket. |
| self-improvement | Development | Mine local agent prompt history for how the skill catalog is really used, then land the changes that usage evidence supports. |
| ticket | Development | Manage work across trackers (Linear, Jira, GitHub, etc.) respecting each platform's native object graph. Drafts durable tickets focused on invariants. |
Add the marketplace:
/plugin marketplace add tony/skillsYou can also browse available plugins with /plugin > Discover.
Then install any plugin by the name in the table above:
/plugin install commit@skillsAdd the marketplace:
codex plugin marketplace add tony/skillsThen install any plugin by the name in the table above:
codex plugin add commit@skillsEach workflow is one skill, invoked as pr:deslop under Codex and
/pr:deslop under Claude Code.
All plugins are language-agnostic. Instead of hardcoding language-specific
tools (e.g., pytest, jest, cargo test), they dynamically read project
conventions from AGENTS.md or CLAUDE.md to discover:
- Test suite, linter, formatter, and type checker commands
- Commit message formats
- Test patterns
This ensures compatibility across languages like Python, TypeScript, Rust, and Go.
Scripts use uv to manage Python dependencies.
Install uv:
curl -LsSf https://astral.sh/uv/install.sh | shor:
wget -qO- https://astral.sh/uv/install.sh | shSee uv installation docs for other methods.
uv run ./scripts/marketplace.py lintDry-run:
uv run ./scripts/marketplace.py syncWrite changes to marketplace.json:
uv run ./scripts/marketplace.py sync --writeuv run ./scripts/marketplace.py check-outdated.agents/skills/ is a generated, flattened mirror of every plugin skill for
agents outside Claude Code (Cursor, pi, Antigravity, Grok). It contains
spec-only frontmatter with no host-specific inline-bash expansion. Files are
copied and links rewritten to make each exported skill self-contained.
Codex reads skills/ in place and consumes plugins directly via
.codex-plugin/plugin.json.
uv run ./scripts/marketplace.py portableVerify the committed tree matches the generated plugins:
uv run ./scripts/marketplace.py portable --check- Edit:
plugins/*/skills/ - Do not edit generated files:
.agents/skills/,.agents/plugins/, orplugins/*/.codex-plugin/
.agents/portable-manifest.json tracks exported skill sources and
bundled files.
Hosts scanning .agents/skills/ read the tree directly. The skills CLI
also reads this tree:
npx skills add tony/skillsThis installs only the export. The metadata.pluginRoot in the marketplace
manifest prevents duplicate skill renderings from being exposed.
Lint:
uv run ruff check ./scripts/Format check:
uv run ruff format --check ./scripts/Type check:
uv run basedpyright ./scripts/See the official Claude Code plugin docs for authoring guides, component schemas, and marketplace publishing.
MIT