agent-skills

Set up Microsoft SkillOpt for Agent Skills improvement. Use when installing SkillOpt, preparing per-skill train/val/test eval splits, configuring Codex CLI or provider-backed execution, running SkillOpt against SKILL.md bodies, reviewing best_skill.md, or importing optimized skill changes. Do not use for ordinary skill authoring or manual repo validation.

Create, draw, generate, edit, verify, and export draw.io/diagrams.net `.drawio` diagrams. Use when the user asks for editable diagrams, flowcharts, architecture, sequence, ER/UML/state, BPMN, SysML, ML/DL, swimlane, timeline, network, icon-rich technical diagrams, or PNG/SVG/PDF exports; do not use for charts/plots or artistic image generation.

Write a compact handoff document so a fresh agent, thread, or session can continue. Use when context is long, the user wants to switch threads, work must be paused, or the agent needs to summarize current state without duplicating existing docs or diffs.

Interview fuzzy Claude Code coding requests into user-verified implementation specs with source challenge, ADR gate, validation plan, and execution prompt. Use when the user asks for /claude-spec-interviewer, a Claude Code implementation plan, PRD, requirements, plan before coding, Plan Mode planning, persisted spec, ADR decision, CLAUDE.md evidence, or .claude/rules evidence. Do not use for direct implementation, pure brainstorming, memory cleanup, or authoring Claude instruction files only.

Interview fuzzy Cursor Agent coding requests into user-verified, Cursor-ready implementation specs. Use when the user asks for a spec, implementation plan, PRD, requirements, or a plan before coding for a feature, bugfix, refactor, migration, repo-wide change, or architecture task, or needs acceptance criteria, validation commands, ADR decisions, or rollout notes. Saves the spec and a Cursor execution prompt. Do not use for fully specified tasks or direct implementation requests.

Create, transform, or review animated logo asset pipelines for GitHub READMEs, including starter asset creation when no logo exists, transparent-background transformations, SVG/Lottie source choices, README-safe picture markup, reduced-motion fallbacks, GitHub Pages demos, and asset validation. Use when working on README logos, profile README hero images, animated repository branding, or README image compatibility.

Set up ADR governance or refactor/review repositories from explicit ADRs, stack rules, code examples, or a requested new-repo adoption plan. Use when a repo needs agent-facing ADR guardrails or when existing/new code must follow documented source structure, runtime boundaries, request patterns, backend composition, env handling, and validation rules. Do not use for tiny edits, style-only cleanup, or no-evidence tasks that do not request governance adoption.

Set up and use CodeGraph with ast-grep for semantic repository exploration, impact analysis, structural search, rule testing, and approval-gated refactors. Use when a repo needs CodeGraph CLI/MCP configuration for Codex, Claude Code, Cursor, or another client; when the user asks about callers, call paths, affected files, AST patterns, sgconfig, analysis-tool updates, or safe codemods; or when syntax evidence should confirm semantic findings.

Interview, source-challenge, verify, save, and ADR-gate fuzzy coding requests into Codex-ready implementation specs. Use when a feature, bugfix, refactor, migration, repo-wide change, or architecture task needs user-verified requirements, source-backed decisions, durable architecture decisions, acceptance criteria, validation commands, rollout notes, saved spec/ADR files, and a Codex execution prompt. Do not use when already fully specified or when the user wants direct implementation now.

Audit, review, clean up, and prune Claude Code durable context. Use when the user asks about CLAUDE.md, CLAUDE.local.md, .claude/rules, user Claude rules, Claude Code auto memory, /memory, stale instructions, memory pollution, sensitive context, settings such as autoMemoryEnabled or claudeMdExcludes, or where a Claude instruction should live. Do not use for Codex memory, Cursor rules, Claude app memory, Anthropic API Memory Stores, or generic docs cleanup.

Audit, review, clean up, and prune Codex memories. Use when the user asks about ~/.codex/memories, stale or noisy memories, memory pollution, cross-repo rule leakage, sensitive memory contents, memory config tuning, cleanup plans, or whether entries belong in memory, AGENTS.md, repo docs, skills, config, or deletion. Do not use for ordinary repo docs cleanup.

Audit, review, clean up, and prune Cursor durable context. Use when the user asks about Cursor rules, .cursor/rules, .cursorrules, User Rules, Team Rules, Cursor memory-bank artifacts, stale or noisy persistent context, rule conflicts, sensitive rule contents, or why Cursor keeps forgetting or using old guidance. Do not use for Codex memory cleanup, ordinary repo docs cleanup, generic prompt engineering, or direct implementation work without Cursor persistent context.

Maintain a public Agent Skills repository. Use when updating skill catalogs, validating SKILL.md files, preparing a release, deprecating skills, reviewing install commands, checking README consistency, or keeping a skills repo ready for npx skills installation.

Prevent Codex context-window exhaustion during long-running refactors, repo audits, migrations, debugging sessions, or tasks with large logs, many file reads, or repeated tool output. Use when Codex context is getting high, /compact may be needed, or the user asks for context-efficient workflows.

Diagnose and fix bugs through reproduction, minimization, hypotheses, instrumentation, targeted fixes, and regression coverage. Use when the user reports failing behavior, a broken test, runtime error, flaky workflow, or asks for debugging before implementation.

Convert a spec, roadmap item, or rough plan into independently implementable issue slices with scope, dependencies, acceptance criteria, and validation. Use when the user asks to split work into GitHub issues, local markdown tasks, or agent-ready implementation chunks.

Write a concise PRD or implementation brief from current context, separating problem, solution, user stories, decisions, testing, rollout, and out of scope. Use when the user asks for a PRD, product brief, implementation brief, or spec handoff.

Build or plan an explicitly throwaway prototype to answer a design, API, UX, performance, or integration question, then capture findings and cleanup. Use when the user asks for a spike, prototype, proof of concept, or experiment before committing to production design.

Map unfamiliar code areas before implementation by identifying entry points, ownership boundaries, callers, data flow, ADRs, validation, and likely edit surfaces. Use when the user asks to understand a subsystem, zoom out, map a repo area, or prepare for a risky change.

Implement behavior with a red-green-refactor loop, observable tests, and small vertical slices. Use when the user asks for TDD, test-first work, regression-first fixes, or a feature that needs clear behavior before implementation.

Relentlessly question a plan, design, migration, release, or repo maintenance proposal until assumptions, risks, constraints, and decision branches are clear. Use when the user wants to be grilled before implementation or wants a plan challenged.

Create short Architecture Decision Records under docs/adr. Use when the user makes a repo-level decision, changes skill format, publishing, validation, license, layout, or maintainer policy, and wants a concise ADR without documentation bloat.

Bootstrap repo-local agent context for Codex and other coding agents. Use when setting up AGENTS.md, docs/agents, issue tracker rules, triage labels, domain docs, ADR locations, validation commands, or skill usage instructions for a repository.

Diagnose failed CI jobs and build pipelines using small-log, evidence-first debugging. Use when the user shares a failed GitHub Actions, GitLab CI, Vercel, or package-manager build and wants root cause, minimal fix, and regression test guidance.

Review dependency updates, package manager changes, lockfile diffs, version bumps, and dependency risk. Use when the user asks whether dependency updates are safe, wants to upgrade packages, or needs a dependency PR reviewed.

Audit repository documentation, README, examples, installation steps, onboarding flow, API docs, and contribution docs. Use when the user asks whether docs are clear, complete, current, or ready for a public repo.

Triage repository issues through a maintainer decision tree. Use when the user asks to classify issues, apply labels, identify missing information, mark ready-for-agent work, split vague reports, or clean up an issue inbox.

Review pull requests for correctness, maintainability, tests, security, docs impact, release risk, and agent-induced failure modes. Use when the user asks for a PR review, diff review, merge readiness check, or maintainer feedback.

Prepare repository releases. Use when the user asks to draft release notes, update changelog, check semver, verify CI, create a release checklist, or decide whether a public repo is ready to tag and publish.

Audit repository health and maintenance readiness. Use when the user asks for a repo review, maintainer audit, cleanup plan, onboarding audit, technical debt scan, CI/docs/release hygiene review, or public repo readiness check.

Review public repository security hygiene at a baseline maintainer level. Use when the user asks for secret-leak checks, SECURITY.md review, dependency hygiene, GitHub settings checklist, CI safety, or public release security readiness.

Create or review Agent Skills with correct SKILL.md frontmatter, trigger descriptions, concise workflow design, progressive disclosure, references, scripts, assets, safety rules, and installability. Use when writing a new skill, improving an existing skill, or validating a public skills repo.

Help install, list, update, remove, or troubleshoot Agent Skills using the Vercel skills CLI. Use when the user asks about npx skills add, installing a whole repo, installing one skill, global vs project installs, symlink vs copy, or Codex/Claude/Cursor agent targets.