tony
UserLanguage-agnostic AI workflow plugins for Claude Code and Codex. Every workflow is a skill.
Categories
Indexed Skills (135)
changelog-rewrite-aggressively
Rebase out the branch's earlier changelog commits and regenerate its entries fresh; commits only with --commit
pr-rewrite
Rewrite an existing PR description from scratch against the branch's current net change, carrying forward context that still matters
agy
Delegate a task to Google's Gemini via the Antigravity (agy) CLI. Use this skill when the user explicitly asks to use Antigravity, agy, Gemini, or Google's model for a task, or when you determine that Gemini would provide better results for a specific task. Detects the agy binary, falls back to the gemini CLI, then to agent --model gemini-3.1-pro if neither is available.
codex
Delegate a task to OpenAI's GPT via the Codex CLI. Use this skill when the user explicitly asks to use Codex, GPT, or OpenAI for a task, or when you determine that GPT would provide better results for a specific task (e.g., tasks requiring OpenAI-specific strengths). Detects the codex binary, falls back to agent --model gpt-5.4-high if unavailable.
cursor
Delegate a task to Cursor's agent CLI. Use this skill when the user explicitly asks to use Cursor or the agent CLI for a task, or when you determine that Cursor's agent would provide better results for a specific task. Requires the agent binary — there is no fallback for this skill.
study-deps
Clone and study your project's dependencies at the exact versions you use. Scans manifest files (package.json, pyproject.toml, Cargo.toml, go.mod, etc.), resolves official source repositories, clones them to ~/study/<language>/, and creates version-pinned git worktrees. Use when the user wants to read upstream source code, understand how a dependency works, or study a library at the exact version their project depends on.
xfail
TDD xfail workflow with hermetic reproduction guards — reproduce a bug as a strictly-expected-to-fail test, verify the reproduction is genuine, apply a fix with diff-gate enforcement, and remove the xfail marker. Use when fixing bugs that need strict proof the test fails for the right reason, not because a mock is misconfigured. Triggers on "xfail", "strict xfail", "hermetic TDD", "reproduce with xfail", or "expected failure workflow".
action-worktree
Use when taking one ticket or a related ticket group into its own branch and git worktree for implementation.
action-worktrees
Use when fanning several tickets into parallel branches and git worktrees, with related tickets grouped together.
business-report-leadership
Use when rendering a tier-0 leadership report from a completed business-value run, with SCQA and a conservative value case.
business-research
Use when collecting evidence for the business value of an AI skill or workflow into a provenance-tagged research package.
double-check-align
Use when prior verification was framed as a revision log and the user wants a clean, standalone restatement.
github-actions-update-actions
Use when updating every outdated GitHub Action across one repository or a fleet, one researched change at a time.
lean-tighten
Use when tightening named files or pasted text in place by removing verbose prose, brittle references, and low-value noise.
package-updater-update-package
Use when updating one named dependency to a target version everywhere it is pinned across one repository or a fleet.
package-updater-update-toolchain
Use when updating one runtime or toolchain pin across version files, package metadata, CI, and other repositories.
package-updater-update
Use when bringing every outdated dependency and toolchain pin current across one repository or a fleet.
pr-deslop
Use when cleaning AI slop, verbose commit messages, brittle references, or low-value changes from a branch before review.
pytest-optimizer-00-scan
Use when starting or re-baselining a pytest optimization pass by profiling tests and fixtures without editing the suite.
pytest-optimizer-01-benchmark
Use when benchmarking pytest optimization hypotheses from 00-scan in isolation to identify speedups that beat measured noise.
pytest-optimizer-02-plan
Use when ranking validated pytest speedups from 01-benchmark into an approved, ordered, commit-by-commit plan.
pytest-optimizer-03-execute
Use when applying an approved pytest optimization plan as separate verified commits with resumable progress.
changelog-refresh
Update the branch's own changelog entries to match its current net change; commits only with --commit
changelog
Generate CHANGES entries from branch commits and PR context
merge-pr-this
Merge one PR via gh with a merge commit matching the repo's git history — readiness-gated, CI-watched, trunk synced after
pr-refresh
Refresh an existing PR description to match the branch's current net change, preserving structure and customizations
business-case-study-internal
Write the tier-2 internal case study — situation, what was built, tagged outcomes with denominators, lessons, replication guide for other teams
business-case-study-public
Write the tier-3 public case study — hard sanitization, every headline claim triangulated against external evidence, candid limitations mandatory
business-pr-release
Write the tier-3 one-page announcement — figures drawn only from the public case study, denominators attached, limitations one-liner included
business-report-org-wide
Render the tier-1 org-wide projection — explicit adoption and realization inputs, scenario spread, sensitivity ranking, plain-language company close
changelog-recut
Rebase out the branch's earlier changelog commits and regenerate its entries fresh; commits only with --commit
disk-reclaim
Reclaim disk space through a proof-gated plan — clears regenerable caches, merges proved-redundant copies, and protects agent history
disk-usage
Survey disk usage across every filesystem layer and classify what is reclaimable, without deleting anything
github-actions-update-action
Update one named GitHub Action to its current version — verify the tag exists, research the upgrade, and commit it with release links
merge-pr-multiple
Merge a set of PRs one at a time — detect stack vs independent set, rebase and resolve conflicts between merges, watch CI, merge each via gh
pr-merge-commit
Generate a gold-standard merge commit message from branch diff
pr-recut
Rewrite an existing PR description from scratch against the branch's current net change, carrying forward context that still matters
pr-review-pr
Review a PR description against gold-standard patterns
pr
Generate a gold-standard pull request description from branch diff
rebase
Rebase current branch onto trunk (origin/master or origin/main), predict and resolve conflicts
release-bump
Discover what version comes next — enumerate candidates from the project's own scheme (a1→a2, 0.1.9→0.1.10, 0.2.0, 0.2.0a0), confirm with the user, then cut the release with the same safe defaults as the `release-cut` skill.
commit
Create a git commit following project conventions
gemini
Use when running a prompt through Google's Gemini directly. A manual invocation point for /model-cli:agy — the Antigravity (agy) CLI supersedes the standalone gemini CLI (Google retires it on 2026-06-18), so both entry points resolve through the same backend chain: agy, then the gemini CLI, then agent --model gemini-3.1-pro.
model-cli-gemini
Use when invoking Google's Gemini directly through the shared Antigravity, Gemini CLI, or agent fallback chain.
model-cli-gpt
Use when invoking OpenAI GPT directly through the same Codex CLI or agent fallback used by the codex skill.
lean-trim-comments
Use when existing source comments or docstrings are bloated, dense, repetitive, AI slop, or should be trimmed, debloated, and kept light.
agent-messaging-session-relay
Use when sending a message to another Codex/Claude session or telling an agent in another tmux pane a status, finding, instruction, or schema change.
tailwind-spacing-audit
This skill should be used when the user mentions uneven spacing, alignment issues, inconsistent gaps between elements, mixed margin/gap strategies, navbar or toolbar spacing problems, or asks to "clean up", "normalize", or "audit" spacing in a Tailwind component. This skill also applies when reviewing UI code where flex/grid containers use a mix of gap-*, margin (m*, me-*, ms-*), padding, and fixed widths to space sibling elements — even if the user doesn't explicitly say "spacing." If someone pastes a nav bar, toolbar, header, or action bar and says "something looks off," this skill is almost certainly what they need.
serene-bliss
Use when the user wants to brainstorm and refine developer-experience, documentation, or tooling UX work through a "serene DX" aesthetic lens — three fixed variants that sweep across DX Bliss (frictionless), DX Serenity (calm clarity), and DX Sublimity (showcase-grade novelty). Triggers on phrases like "serene bliss", "DX bliss", "DX serenity", "DX sublimity", "reader happiness", "make this serene", "serene developer experience", or "serene DX". Runs the /weave:serene-bliss command, which dispatches three lens variants across all available models in parallel and judges each refine pass with a multi-model panel (Claude + Antigravity + GPT, peer-only averaging).
release-cut
Cut a release at an explicit version — bump version files, refresh the lockfile, date CHANGES/MIGRATION, and commit. Never pushes or tags unless explicitly flagged.
release-update-downstream-packages
Roll a newly published package release out to every consumer repo you maintain — discover consumers under your workspace roots, bump pins, re-lock, commit, push, and verify CI with gh.
research-study-deps
Clone and study your project's dependencies at the exact versions you use. Scans manifest files (package.json, pyproject.toml, Cargo.toml, go.mod, etc.), resolves official source repositories, clones them to ~/study/<language>/, and creates version-pinned git worktrees. Use when the user wants to read upstream source code, understand how a dependency works, or study a library at the exact version their project depends on.
review-address
Use when the user wants to act on code-review feedback on the current branch — review comments, findings, PR feedback, or a reviewer's punch list. Triggers on phrases like "address the review items", "respond to the review", "fix the review findings", "handle the PR comments", "action the reviewer feedback", or "address each issue in separate commits". Fixes only what the branch introduced, one finding per commit, behind the project's quality gates; anything pre-existing or history-rewriting is surfaced for a decision instead of silently done.
slop-scan
Use when the user wants to scan the current repo for AI slop, verbosity, fragile hard-coded references (line numbers, test counts, file counts), or low-value contributions in tracked files, with each finding landing as its own atomic forward-going commit. Triggers on phrases like "scan for slop", "audit repo for slop", "deslop the repo", "remove slop from this codebase", "scrub the repo", "scan the codebase for AI signatures", or "clean up slop without rewriting history". Does NOT rewrite history; every finding lands as a forward-going commit, with the project's formatter, linter, and type-checker running before each commit.
spike-bakeoff
Use when the user wants to try 2–4 different strategies for the same goal in parallel and pick a winner — a bakeoff, arena, tournament, gauntlet, shootout, or head-to-head where each contender is built for real in its own git worktree, adversarially judged, and none of it lands as commits. Triggers on phrases like "bakeoff", "bake off two approaches", "arena them", "run the gauntlet", "hold a tournament", "fan out different approaches", "try it three different ways", "competing implementations", "head-to-head", or "which approach wins". Varies the strategy, not the model — for running one prompt across different AI models (Claude, Antigravity, GPT), use the weave plugin instead. Ends with every contender stashed with a recovery SHA, a judged verdict, and a commit-by-commit plan to land the winner through the project's quality gates.
spike-probe
Use when the user wants to prove out a feature or fix fast before committing to an implementation — a spike, probe, blitz, sprint, bolt, speedrun, MVP pass, or proof-of-concept that mutates the working tree but must not land as commits. Triggers on phrases like "probe it", "quick probe", "spike into", "do a spike", "blitz it", "bolt through it", "speedrun the fix", "do a sprint to handle this without committing", "take a stab at it", "prove it works, then plan", "MVP this then clean it up", "probe those review items", or "get it working before my meeting". The goal may be typed or inferred from conversation context (review findings, a failing test under discussion). Ends with the working tree stashed and a commit-by-commit plan to land the work through the project's quality gates.
tdd-fix
TDD bug-fix workflow — reproduce a bug as a failing test, find root cause, fix, and verify
tdd-xfail
TDD xfail workflow with hermetic reproduction guards — reproduce a bug as a strictly-expected-to-fail test, verify the reproduction is genuine, apply a fix with diff-gate enforcement, and remove the xfail marker. Use when fixing bugs that need strict proof the test fails for the right reason, not because a mock is misconfigured. Triggers on "xfail", "strict xfail", "hermetic TDD", "reproduce with xfail", or "expected failure workflow".
weave-architecture
Weave architecture — generate project scaffolding, conventions, skills, and architectural docs across Claude, Antigravity, and GPT, then synthesize the best architecture
weave-ask
Weave question — ask Claude, Antigravity, and GPT the same question in parallel, then synthesize the best answer
weave-brainstorm-and-refine
Use when the user wants to generate multiple original ideas from different AI models then iteratively judge, weave, and refine them into the best possible result across multiple passes. Triggers on phrases like "brainstorm and refine", "generate ideas then improve", "explore options then synthesize", or "brainstorm then polish"
weave-brainstorm
Use when the user wants multiple independent ideas, alternatives, or approaches from different AI models for a creative prompt, design question, or open-ended problem. Triggers on phrases like "brainstorm", "give me ideas", "multiple approaches", "what are my options", or "explore alternatives"
ruff-bump
Move one repo or a whole fleet onto a new ruff release — predict which rules can fire against each repo's own select list, gate on the resolver seeing the version, then land one reviewed commit per rule
situate-situational-awareness
Use when a session starts on unfamiliar or resumed work — when asked to catch up, get oriented, or figure out where things left off, what a branch is doing, what a pull request or its review threads are asking for, or what the state of this repository is before touching it.
situate
Gain situational awareness — read the branch, its diff, its PR, its tickets, and the project's own conventions, and report where the work stands
situate-brief
Use when the user signals confusion without naming a subject — "huh", "huh?", "wait, what?", "sorry, you lost me", "no idea", "slow down", "explain simply", "keep it short". Answers in five lines or less, followed by numbered single-line options. Not for a question that names its own subject.
situate-refocus
Re-derive what this work is for from its ticket and pull request, sort every commit into on-goal, load-bearing, and drift, name what the goal asked for that is still missing, and propose the correction
situate-what
Say what is going on in five lines or less — the session, the branch, the pull request and ticket if they exist — with numbered options when there is a real choice to make
disk
Use when disk space is low; find large directories, free space, decide if Claude sessions or npm cache are safe to remove, or shrink WSL.
double-check
Use when asked “are you sure/certain” or “I don't trust that”; recheck code/files from source and repeat prior analysis in full, not just changes.
gh-create-issue
Use when filing a GitHub issue for a bug, feature, audit, review finding, or investigated piece of work.
gh-rendered-markdown
Use when Markdown must render: never hard-wrap issue-body paragraphs; fix GitHub code fences, fold logs into details, and pin source links to tags.
git-branch-redo-from-scratch
Use when a branch or proof of concept should be redone from scratch: throw away the approach, start implementation over, and treat tests as the spec.
git-branch-soft-reset-and-recommit
Use when commit history must be rebuilt into atomic reviewable commits while preserving the branch's final tree and authorship.
github-actions-bumping-github-actions
Use when workflow actions are out of date across repos, Dependabot opens action-bump PRs, or pinned actions/checkout needs its latest release.
model-cli-agy
Use when delegating to Google Gemini or Antigravity for a second opinion from Google's model; falls back through Gemini and agent CLIs.
model-cli-codex
Use when asked to use Codex, GPT, OpenAI's model, or the Codex CLI to delegate a task; falls back to the GPT agent when needed.
model-cli-cursor
Use when asked to use Cursor, Cursor's CLI, or the agent binary to delegate a prompt or task; no fallback is available.
package-updater-updating-packages
Use when packages are out of date across repos; bump dependencies, refresh uv.lock, update .tool-versions/packageManager pins, and commit each.
pytest-optimizer
Use when a pytest suite or its fixtures are slow and need measured, safety-gated optimization across the full pipeline.
gpt
Use when running a prompt through OpenAI GPT directly. An alternative invocation point for /model-cli:codex — both use the same backend (codex binary or agent --model gpt-5.4-high fallback). Use this or /model-cli:codex interchangeably.
brainstorm-and-refine
Use when the user wants to generate multiple original ideas from different AI models then iteratively judge, weave, and refine them into the best possible result across multiple passes. Triggers on phrases like "brainstorm and refine", "generate ideas then improve", "explore options then synthesize", or "brainstorm then polish"
brainstorm
Use when the user wants multiple independent ideas, alternatives, or approaches from different AI models for a creative prompt, design question, or open-ended problem. Triggers on phrases like "brainstorm", "give me ideas", "multiple approaches", "what are my options", or "explore alternatives"
refine
Use when the user has an existing draft, text, code, or artifact and wants it iteratively improved through multi-model critique and weaving across multiple passes. Triggers on phrases like "refine this", "improve this", "make this better", "iterate on this", or "polish this"
lean-writing
Use when commit messages or prose must stay tight: cut filler, padding, preamble, AI fluff, and stories of attempts; lead with current results.
disk-reclaim
Reclaim disk space through a proof-gated plan — clears regenerable caches, merges proved-redundant copies, and protects agent history
disk-usage
Survey disk usage across every filesystem layer and classify what is reclaimable, without deleting anything
disk
Use when a machine is low on disk space, when the user wants to find what is consuming it, or when the user asks whether a large directory is safe to delete. Triggers on "disk is full", "out of space", "what is eating my disk", "where did my space go", "clean up my drive", "find the biggest directories", "reclaim disk space", "clean up crap", "my home directory is huge", or "free up space". Also on safe-to-delete questions about specific consumers — "can I delete my npm cache", "are my old agent sessions safe to remove", "is this backup a duplicate", "node_modules is eating my SSD" — and on virtual-disk pressure like "why is my ext4.vhdx so large", "WSL is taking hundreds of gigabytes", or "shrink my WSL disk". Surveys every filesystem layer, classifies each candidate as regenerable cache, proved-redundant copy, or irreplaceable agent history, and reclaims only what a proof says is safe. Protects LLM transcripts and session stores by default. Never halts a VM or WSL guest on its own initiative.
package-updater-update-toolchain
Move runtime and toolchain pins — .tool-versions, .nvmrc, .python-version, packageManager, engines — one tool per commit, each release in the span linked
package-updater-updating-packages
Use when dependencies or toolchain pins are out of date across one repo or a whole fleet — bump dev packages, refresh a lockfile (uv.lock, pnpm-lock.yaml, package-lock.json, Cargo.lock, go.sum), find outdated packages with ncu, uv, pnpm or cargo, hold a package back in .ncurc, or move a .tool-versions, .nvmrc, packageManager or engines pin.
gh-create-issue
Use when filing a GitHub issue — reporting a bug, proposing a feature or a piece of work, or turning an audit, review finding, or investigation into something tracked. Reproduces and gathers evidence before writing prose, checks the repository's own templates and existing issues for a duplicate, pins every source link to a tag or commit, strips local paths and PII, previews the body through GitHub's renderer, and opens it with `gh` only after you approve the full title and body.
gh-rendered-markdown
Use when writing anything a renderer will show a human — a GitHub issue, pull request, discussion or review-comment body, release notes, a tracker ticket, or a repository markdown file. Governs the markup that decides whether it renders and the links that decide whether it survives: never hard-wrap a comment body, backtick every symbol, pin source links to a tag or a 7-character commit, one shell command per fence, long output folded into `<details>`, nested sections instead of tables, and no local paths or PII. Guidance only; it never edits files on its own.
git-branch-redo-from-scratch
Use when a branch's code works but its implementation should be replaced from scratch rather than tidied — a proof of concept that became the real thing, an approach found halfway through that the earlier code does not reflect, or a structure that fights the problem. Studies the branch into a coverage ledger, treats its tests as the specification, reimplements from those requirements instead of copying the old code, then reconciles the result against the ledger entry by entry. The net change may legitimately differ from the original, which is kept as reference and fallback.
git-branch-soft-reset-and-recommit
Use when a commit history needs rebuilding rather than the code it contains — `wip` commits to squash, one commit doing five unrelated things, or a history no reviewer can follow. Collapses everything with a soft reset and rebuilds it as atomic commits in the project's own message format, preserving authorship, proving the resulting tree is byte-identical to what it replaced, and gating each commit through the project's checks. Ships an editor-free interactive rebase toolkit for reordering, squashing, and verifying from an agent shell with no TTY.
package-updater-update-package
Take one named package to a target version everywhere it is pinned — research the span, land one commit per repo with a why/what body and verified links, then its follow-up
package-updater-update
Find every outdated dependency and toolchain pin across one repo or a fleet and bring them current — toolchain, named bumps, bulk refresh and fallout as separate researched commits
action-worktrees
Use when the user wants to fan several tickets out into parallel branches and git worktrees — one worktree per ticket by default, tickets grouped onto a shared branch when they clearly overlap. Triggers on phrases like "set up worktrees for these tickets", "fan out my Linear queue", "work these three issues in parallel", "a worktree per ticket", or "batch these bugs into branches". Discovers and groups tickets strictly read-only, confirms the grouping at a plan gate, then drives each unit through the `action-worktree` skill's procedure — one subagent per worktree where the host supports it, sequential otherwise.
github-actions-bumping-github-actions
Use when GitHub Actions pins are out of date, when dependabot has opened action-bump pull requests, when asked which workflow actions need updating, or when auditing `uses:` versions across one or many repositories.
github-actions-update-action
Update one named GitHub Action to its current version — verify the tag exists, research the upgrade, and commit it with release links
github-actions-update-actions
Audit every GitHub Action across one repo or a whole fleet and update the out-of-date ones — one researched commit per action, then close dependabot's PRs by citing them
Showing top 100 of 135 skills by quality score. Browse the full catalog at /browse.
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.