Agent skills

Task-focused instructions, references, templates, and tools for skills-compatible agents. Each active skill has its own directory and SKILL.md entry point.

Install and use

Read the selected skill's entry point, licence, dependencies, and account requirements. The Skills CLI supports:

npx skills add tristanmanchester/agent-skills

This runs a CLI and changes the selected agent's skill installation. Select the skills relevant to your work, or use your host's supported installation method. Keep each entry point with its relative supporting files and record the source revision. Do not keep duplicate active copies of the same skill.

Start with SKILL.md; load the references, assets, or scripts it identifies when needed. Not every skill has a separate README. Resolve bundled scripts from the installed skill directory, not an unrelated project's scripts/ directory. Instructions do not install dependencies, establish service access, or authorise live mutations. Licences apply per skill; there is no blanket repository licence.

Catalogue

The repository contains 36 active skill directories. The September 2026 overhaul reviewed 37 baseline directories: 33 update PRs and one retirement PR have merged, two skills were retained unchanged, and the write-good-docs packaging proposal (#22) remains closed unmerged. Its existing skill is retained.

meta-ads-control was retired after its useful operations were consolidated into meta-ads-cli. The required merge order, #9 before #10, was verified. There is no replacement compatibility entry point.

Skill / entry point	Purpose
agent-use	Design and audit agent-facing capabilities, contracts, and recovery.
ai-codebase-deep-modules	Design cohesive modules, explicit ownership, and enforceable boundaries.
animating-react-native-expo	Build native animations and gestures with a compatible Expo stack.
assemblyai-transcribe	AssemblyAI transcription workflows and speaker-aware local exports.
audit-openclaw-security	Review OpenClaw exposure, policy, secrets, and diagnostic handling.
auditing-appstore-readiness	Assess source, archive, runtime, and store submission evidence separately.
designing-beautiful-websites	Design and review purposeful, distinctive, accessible websites.
display-quantitative-information	Create and critique faithful quantitative displays.
exa-search	Use explicitly selected Exa search and content-retrieval workflows.
expo-revenuecat-superwall-integration	Integrate purchases, offers, identity, restores, and entitlements.
extracting-mistral-ocr	Mistral document extraction with bounded, organised exports.
fabric-api	Build Fabric.so SDK/API integrations and upload workflows.
fabric-cli	Operate Fabric.so through its native CLI and scoped project memory.
generating-novel-ideas	Explore distinct mechanisms, critique concepts, and define useful tests.
giving-presentations	Develop audience-focused narratives, visuals, and rehearsal plans.
good-services-service-design	Review whole services with explicit evidence and scoped scoring.
ios-simulator	Run and inspect simulator apps with explicit targets and checked outcomes.
jax-development	Develop, debug, profile, and benchmark JAX code.
meta-ads-cli	Use the official Meta Ads CLI with bounded advanced API operations.
nature-article-writer	Prepare scientific manuscripts with journal-specific checks.
nextjs-framer-motion-animations	Implement Motion interactions and appropriate Next.js transition boundaries.
optimising-expo-react-native-performance	Measure and improve native performance against reproducible baselines.
parallel-ai-search	Use explicitly selected Parallel search, research, and bounded job workflows.
react-native-skia	Build custom native graphics, gestures, snapshots, and shaders.
reddit	Read-only research through permitted web/connector access or an approved OAuth API client; no anonymous JSON client.
relationship-science-coach	Practical relationship coaching with contextual safety and calibrated evidence.
resend-api	Build Resend SDK/API integrations, verified events, and Automations.
resend-cli	Operate Resend through its native CLI with deliberate send and campaign scope.
rust-anti-slop	Apply the repository's intentional Rust code and lint policy.
styling-nativewind-v4-expo	Configure and use the explicitly versioned NativeWind v4 stack.
textual-tui	Build and test Textual interfaces, workers, and starter applications.
todoist-api	Use Todoist's official CLI/API with explicit bulk-write recovery.
track17	Track parcels and ingest authenticated 17TRACK updates.
tracking-pettracer-location	Inspect owner-authorised PetTracer fixes and history through a bounded unofficial reader.
wordly-wisdom	Analyse decisions with explicit assumptions and checked arithmetic.
write-good-docs	Write and edit evidence-preserving technical documentation; packaging proposal #22 remains closed unmerged.

Review records

The September 14 PR review records the final skill merges, checked bot findings, local tests, and remaining validation limits. The machine-readable checkpoint records all baseline dispositions. GitHub remains the authority for later PR state.

The September 13 source review and early September 14 template follow-up are historical records. Their open-PR counts describe those earlier checkpoints, not the current repository. Merging source does not update existing installations or establish live-provider, native-device, or agent-output performance.

Contributing

Follow the Agent Skills format: precise activation, task-focused instructions, and supporting material loaded when useful. Preserve useful methods and verified helpers. Check changing contracts against current primary sources, and record which tests actually ran.

Keep one skill's changes in one PR. Repository-wide catalogue and maintenance changes belong in a separate PR. A fixture, syntax check, mocked test, or merged PR is not evidence that a live integration or a representative agent task passed.