Claude Skills by dzhng

Use the local Codex CLI as an independent second agent. Two branches — (1) proactively run `codex review` for a second opinion after completing a substantive change, before presenting it as done or committing; (2) delegate a well-defined implementation task via `codex exec`, ONLY when the user explicitly asks for Codex to do it. Also covers how to prompt Codex.

Implement an existing spec through committed passes. Use for long or multi-pass specs that need maintenance checkpoints to periodically clean code, handoffs, priorities, and plan bloat before drift accumulates.

Break large features into independently verifiable, human-reviewable slices under a feature directory in specs/. Use for risky or multi-step feature work that needs upfront questioning, API seams, browser-playable checkpoints, HTML visualizations, screenshot gates, staged implementation plans, recursive fog-of-war reslicing, or proactive research into reference implementations/best practices before slicing. Pairs with your project's verification harness and screenshot gates (the browser check...

Use for questions about how, why, or where behavior works in a repository, including questions that name a function or setting. Start behavioral discovery with jg before broad text searches or git history. When delegating repository discovery, instruct the subagent to start with jg. Jevgrep returns relevant files and source excerpts. For exact symbol definitions, string matches, or filenames, use grep or file search instead.

Audit the choices an implementing agent made, not its diff — a pure decision audit that traces the session's history into a choices ledger, changes no code, and never blocks an unsupervised run. Working code still embeds architecture the user never chose; surface it because future work inherits it. Use when the user wants to review the decisions the AI made on their behalf, before merging or committing AI-implemented work, when integrating a delegated subagent's pass, or when a fix "works" bu...

Audit and prioritize performance work through bounded-work and forward-progress checks. Use when asked to find performance issues, investigate freezes/500s/OOMs, review retries or polling for no-progress loops, rank a performance backlog, or implement the next simple performance fixes.

Run fast, progressive experiments to map tunable parameters and their effects. Use when optimizing code, prompts, configurations, or other artifacts against a goal or benchmark, or when a long research run needs a planned hypothesis queue, focused trials, and durable findings.

Use Claude Code as an independent `claude -p` subagent when the user explicitly asks for Claude, wants a second-agent opinion from Claude, or asks to delegate a well-scoped task to Claude. Supports selecting `--model` and thinking/effort level with defaults of `opus` and `high`.

When a spec under specs/<feature>/ is done shipping (a write-spec build, or any planned task that produced a spec), archive it to specs/done/ and rewrite it from a build-plan into a durable rationale — the why, the principles, the invariants — pointing back to the real code for the how. Use when implementation has landed and the plan no longer matches what shipped, or the user says a feature/spec is finished. Pairs with [write-spec](../write-spec/SKILL.md) (the plan this closes) and [review](...

Review changed code for naming, stale references, unnecessary complexity, and comment quality. Use after completing implementation work, before committing, or when the user asks to review or audit code.

Compare screenshots against the intended design, distinguishing approved references from historical baselines. Use for iterative implementation matching, before/after visual review, objective image telemetry, or checking a lone capture for flat, empty, or badly framed content.

Explain a technical spec or proposed change in plain language: what problem it solves, how the solution works, and every schema change.

Eval and improve a skill against golden cases — run the target skill blind in a fresh, context-free subagent on each example input, grade the artifact against the expected outcome, and let the gaps drive the edits. Use when the user wants to test/eval/improve/harden a skill, says "this skill keeps producing X / keeps missing Y", or hands a skill plus example input→expected-output pairs. Pairs with [write-skills](../write-skills/SKILL.md) (the authoring principles every fix obeys).

Guide the user through a quadrant walk that maps the unknowns of a task — open by listing the known knowns, then work through known unknowns, unknown knowns, and unknown unknowns one stage at a time, ending with a complete four-quadrant map in the user's hands. Use when a request is ambiguous or underspecified, the codebase or domain is unfamiliar, the user will "know it when they see it", a reference implementation must be understood before porting, mid-build deviations from the plan need ca...

Run [implement-spec](../implement-spec/SKILL.md) with Codex doing the implementation passes while you orchestrate, integrate, and review. Use when the user asks to implement a spec with Codex, or to run implement-spec with Codex as the implementer.

Make a showreel-grade motion-graphics launch video, with music, that introduces a CLI, library, or product from its repo. Use when asked for a launch, promo, sizzle, or showreel video, an animated intro for a project, or a video to post on X/social.

Best-practice rulebook for a marketing site's pages, organized by page class — campaign landers, product pages, programmatic SEO trees, editorial. Use when writing or updating a landing/marketing page, wiring a campaign lander to an attribution identifier, changing site-wide nav/footer links, touching sitemap/robots/agent-readable surfaces, or auditing a public site's indexing, reachability, and conversion setup.

Open image shots (screenshots, snapshot baselines, montages, renders) in ONE macOS Preview window so the user can eyeball them. Use when the user asks to open / show / pull up / review shots in Preview or Finder, or right after producing visual artifacts the user should look at before signing off.

Refactor cleanly instead of layering sediment. Use for rename requests, stale or misleading names, changed scope, or when a change reveals duplicated concepts, local adapters, obsolete owners, redundant inputs or guards, compatibility wrappers, parallel abstractions, an over-large module that has accreted many responsibilities, or "just tack this on" pressure in any code area.

Closeout pass on a finished substantive change — sequence refactor-clean (shape), code-review (diff), and write-docs (docs) so all three run, in the order where each feeds the next. Use after implementing a slice or feature and before calling it done or committing; when the user asks for a final review, a cleanup-and-document pass, or to "review everything." For a diff-only audit, reach code-review directly.

Use the unprimed sub agent as a second set of eyes before accepting visual work — MANDATORY before declaring any user-reported visual bug fixed or claiming a visual change verified; primed eyes pass defects fresh eyes catch.

Build AI-powered software with TypeSafe: small units of AI intelligence you can use like programming primitives. Its System One models, including Jev, turn natural language and application state into typed judgments and probabilities that code can combine. Use when a feature needs programmable common sense, when brainstorming what AI could make possible in an app, or when an LLM prompt-and-parse step could become a structured decision. Applications include routing, ranking, extraction, verifi...

Write and edit project docs (README/markdown) as a glossary of principles, not a mirror of the code. Use when creating or revising a README; when a doc enumerates exact scenes, scenarios, helpers, class ids, file lists, or command/flag matrices the code already holds; when trimming narrative or changelog out of a doc; or when deduplicating overlapping docs and wiring a root doc to its sub-docs.

Create or revise agent skills. Use when adding a new skill file, renaming a skill, simplifying an existing skill, improving trigger descriptions, or deciding what belongs in a skill versus references, scripts, assets, or ordinary docs.

Write tests that pin real behavior instead of implementation details, config values, or lucky samples. Use when adding tests for new behavior, writing a regression test, fixing a brittle or flaky test, reviewing a test diff, or when a test breaks after a refactor or config change that didn't change behavior.

Audit whether tests earn their maintenance cost and which suite owns each contract. Use when pruning redundant tests, investigating implementation coupling or test-only production hooks, reviewing the value of proposed coverage, or auditing an entire subsystem. Use write-tests to implement the resulting test changes.

Audit or rewrite AGENTS.md so it holds only lasting principles. Run it only when the user asks for it by name; never invoke it on your own.

Explore UI and visual design with image generation, then iterate the real implementation against the user-selected concept. Use when brainstorming visual directions, offering mockup variants, or implementing an approved generated design; do not stop at the mockups or the first code approximation.

Hand a spec that is mid-implementation to another machine or another agent.

Use the local Codex CLI as an independent second agent. Two branches — (1) proactively run `codex review` for a second opinion after completing a substantive change, before presenting it as done or committing; (2) delegate a well-defined implementation task via `codex exec`, ONLY when the user explicitly asks for Codex to do it. Also covers how to prompt Codex.

Implement an existing spec through committed passes. Use for long or multi-pass specs that need maintenance checkpoints to periodically clean code, handoffs, priorities, and plan bloat before drift accumulates.

Break large features into independently verifiable, human-reviewable slices under a feature directory in specs/. Use for risky or multi-step feature work that needs upfront questioning, API seams, browser-playable checkpoints, HTML visualizations, screenshot gates, staged implementation plans, recursive fog-of-war reslicing, or proactive research into reference implementations/best practices before slicing. Pairs with your project's verification harness and screenshot gates (the browser check...

Architecture rules for GPU rendering with TypeGPU/WebGPU — games, data visualisation, canvases, any app that draws with the GPU. Use when adding or changing passes, shaders, buffers, textures, pipelines, frame orchestration, GPU resource lifetimes, or checks that judge rendered output.

Implement a spec using the harness's goal feature.

Conduct a retrospective on a coding session.

Manage contacts, companies, deals, and pipeline with crm.cli — a headless CLI-first CRM backed by SQLite with a virtual filesystem interface

Reproduce observational-memory bugs (reflection, recall, observer, freshness, eviction) by dumping the live `~/.duet/memory.db` into a fixture and driving an eval against it. Use whenever the agent's memory misbehaves — wrong observations, bad reflections, missing recall, runaway tokens — or whenever the user asks to debug, repro, audit, or tune memory.

Increment the package version, commit the release, create a matching version tag, and push main plus the tag. Use when releasing this repo, publishing the package, bumping the version, tagging a release, or when the user asks to "release", "bump ver", "tag", or "push tags".

Write a live eval for new or changed runner/agent behavior using red/green TDD plus a falsification check that proves the eval fails when the behavior is broken. Use whenever you add or modify behavior that should be covered by an eval, when asked to "write an eval", "add an eval", "cover this with an eval", or after landing a feature that needs end-to-end proof it works.