{
  "markdown": "# Agent Skills\n\nPersonal Agent Skills repository for installation through the open [`skills`](https://www.npmjs.com/package/skills) CLI.\n\n## Install\n\nEach skill installs independently. Use the `--skill` flag with the skill's directory name:\n\n```bash\nnpx skills add dhanesh/agent-skills --skill <skill-name>\n```\n\nFor example:\n\n```bash\nnpx skills add dhanesh/agent-skills --skill agent-ready-rails\nnpx skills add dhanesh/agent-skills --skill ai-migration-operating-model\nnpx skills add dhanesh/agent-skills --skill base-in-reality\nnpx skills add dhanesh/agent-skills --skill bug-autopsy\nnpx skills add dhanesh/agent-skills --skill clean-code\nnpx skills add dhanesh/agent-skills --skill context-hygiene-kit\nnpx skills add dhanesh/agent-skills --skill crafting-self-prompting-loops\nnpx skills add dhanesh/agent-skills --skill feynman-walkthrough\nnpx skills add dhanesh/agent-skills --skill knowledge-gardener\nnpx skills add dhanesh/agent-skills --skill okf-site-kit\nnpx skills add dhanesh/agent-skills --skill repo2skill\nnpx skills add dhanesh/agent-skills --skill security-posture-audit\nnpx skills add dhanesh/agent-skills --skill spec-first-planning\nnpx skills add dhanesh/agent-skills --skill starlight-handbook-kit\nnpx skills add dhanesh/agent-skills --skill tmux-agent-herdr-lite\nnpx skills add dhanesh/agent-skills --skill mockstar-mock\nnpx skills add dhanesh/agent-skills --skill verifier-installer\nnpx skills add dhanesh/agent-skills --skill world-model-ledger\n```\n\nTo install every skill in this repo, omit the `--skill` flag:\n\n```bash\nnpx skills add dhanesh/agent-skills\n```\n\n## Skills\n\n| Skill | Description |\n|-------|-------------|\n| [`agent-ready-rails`](agent-ready-rails/) | Read-only audit scoring how ready a repository is for coding agents — grades six rails (runnable verifiers / the format·build·test feedback loop, green-CI ground truth, a copyable house style, navigable context, scoped trusted tools, and human checkpoints with reversibility) from observed repo evidence and emits a severity-ranked readiness scorecard with a leverage-ordered fix list. Optionally installs the missing rails. Grounded in Spotify's Honk case; audits the engineering *system* agents run inside, not code correctness. |\n| [`ai-migration-operating-model`](ai-migration-operating-model/) | Run a code migration (language port, framework/SDK upgrade, service rewrite) as an operating model — rulebook + queues + reviewers + objective verification — instead of a pile of agent edits. Builds the judge first (golden scenarios + a shipped parity differ), then a lint-clean Migration Control Pack (rulebook, dependency map, gap inventory, portability test plan, parity script, work queue, reviewer prompts, phase gates). Four standalone invocations mirror the model's parts: `economics`, `judge`, `pack`, `qualify` — and \"don't migrate yet\" is a first-class verdict. Complements `spec-first-planning` and `crafting-self-prompting-loops`. |\n| [`base-in-reality`](base-in-reality/) | Read-only, research-grounded repo audit — extracts falsifiable claims across algorithm/architecture/business-logic layers, routes each to authoritative sources (arxiv, PubMed, Scholar, JSTOR, OpenAlex, Crossref, Semantic Scholar + NIST/RFC/OWASP/ISO/regulators), verifies against fetched evidence, and adversarially refutes before emitting a severity-graded cited report. No fabricated citations: ungrounded claims are reported as `UNCONFIRMED`. |\n| [`clean-code`](clean-code/) | Applies Robert C. Martin's Clean Code, Clean Architecture and Clean Craftsmanship principles when writing, reviewing, refactoring or designing software — meaningful names, small single-purpose functions, SOLID, simple design, component cohesion and the Dependency Rule, testability, concurrency and cross-cutting concerns. Triggers on ordinary phrasing (\"review this PR\", \"this class is doing too much\") rather than requiring the term, and right-sizes: it pushes the simplest design that fits instead of imposing layers everywhere. Progressive disclosure — an always-on L1 with eight L2 references loaded per decision. |\n| [`context-hygiene-kit`](context-hygiene-kit/) | Install a bounded, scored, tiered context cache into a Claude Code project — keeps working memory lean (anti-bloat) and rot-proof across compactions (anti-rot) via a stdlib Python ledger, a deterministic per-turn transcript harvester, and Stop/PreCompact/SessionStart hooks. Replaces lossy local-model session-summarisers with a lossless one; ships with a 27-test install gate. |\n| [`crafting-self-prompting-loops`](crafting-self-prompting-loops/) | Design and build sound self-prompting loops — draft→critique→revise, autonomous task loops, multi-agent orchestration, and human-checkpointed loops. Produces a filled loop spec, a runnable scaffold, and bakes in the mandatory safety properties (hard-stop backstop, trusted/untrusted two-channel boundary, and a typed loop boundary validating carried state before it feeds the next iteration). |\n| [`feynman-walkthrough`](feynman-walkthrough/) | Understanding-first walkthrough guide (formerly `desirable-difficulty`) — walks a learner through any subject, codebase, or whitepaper with Feynman-grade simplicity and research-backed structure (map → segments → concrete examples), checks understanding only *after* the walkthrough to steer re-coaching, and delivers a standalone reference explainer persisted as a Google Open Knowledge Format (OKF v0.1) bundle with source fingerprints pinned, so later sessions can detect drift and refresh diff-aware. Spaced-retrieval recall tooling included but strictly opt-in. |\n| [`okf-site-kit`](okf-site-kit/) | Turn any Open Knowledge Format (OKF v0.1) bundle into a beautiful browsable static website — a stdlib generator emits a complete Astro + Starlight project with a hero landing page, per-concept pages carrying OKF metadata panels (type badge, tags, resource, producer keys), rewritten internal links, a changelog from `log.md`, and full-text search. Covers every bundle dialect published so far (Google's ga4/stackoverflow/crypto_bitcoin samples, spec-canonical, producer-extended) with graceful `WARN:`-reported degradation; deploy-ready for GitHub Pages. |\n| [`starlight-handbook-kit`](starlight-handbook-kit/) | Scaffold or extend a decision-oriented Astro + Starlight documentation handbook — every topic follows a fixed nine-section skeleton, diagrams and widgets render without JavaScript, and nine CI gates fail the build the moment a page drifts from the contract, making agent-authored content safe. |\n| [`tmux-agent-herdr-lite`](tmux-agent-herdr-lite/) | tmux cockpit with Zellij-like human ergonomics and Herdr-like agent supervision — menus, dashboard, pane navigation, blocked/working/done status detection, and jump-to-status navigation. |\n| [`mockstar-mock`](mockstar-mock/) | Generate a runnable mockstar mock server from a service's specs/docs — OpenAPI, Postman, HAR, curl, GraphQL, and prose docs (md/pdf/docx/url) — normalized into one Endpoint Inventory, full-fidelity (scenarios/handlers/webhooks), Tier 2-enhanced, boot-and-smoke verified, with a provenance + coverage report. |\n| [`repo2skill`](repo2skill/) | The skill-authoring skill: scaffolds a new gate-passing Agent Skill (standard frontmatter, PP-conformant SKILL.md skeleton, README, test stub, outcome-eval stub) and walks the semantic review checklist the mechanical gates can't judge. The source of this repo's vendored gate scripts, now shipped. |\n| [`verifier-installer`](verifier-installer/) | Action-taking sibling of `agent-ready-rails`: detects a repo's stack(s) with a deterministic collector, then installs the missing runnable-verifier loop — format/build/test commands plus a CI workflow that runs them — and proves the loop red→green before handing back. |\n| [`knowledge-gardener`](knowledge-gardener/) | Completes the knowledge trilogy (`feynman-walkthrough` creates, `okf-site-kit` publishes, this maintains): sweeps Open Knowledge Format bundles, reports per-subject FRESH/STALE/UNKNOWN drift against pinned source fingerprints, drives diff-aware refreshes and re-pins, and regenerates any published site. |\n| [`bug-autopsy`](bug-autopsy/) | `feynman-walkthrough`'s sibling for failures: reconstructs a defect end-to-end (evidence-cited timeline, blameless ≥3-deep 5-Whys root cause), writes a lint-checked post-mortem, and persists it into the same OKF knowledge bundle so the failure teaches the next engineer. |\n| [`security-posture-audit`](security-posture-audit/) | Read-only, offline security *hygiene* audit — dependency pinning, committed env/key files, debug/permissive flags, insecure transports, risky CI patterns — severity-graded with file:line evidence and honest not-covered boundaries. Not a CVE scanner or SAST; completes the trust family alongside `scan-leaks` and `base-in-reality`. |\n| [`spec-first-planning`](spec-first-planning/) | Fills the plan band: turns a fuzzy feature request into a lint-clean spec of numbered, testable requirements, then derives a task plan where every task names the check that proves it done — with a total requirement↔task coverage map before handoff to an implementer or a `crafting-self-prompting-loops` loop. |\n| [`world-model-ledger`](world-model-ledger/) | Install a persistent, SQLite-backed world model for a coding agent — entities (symbols/files/modules/real-world referents), interactions, and constraints, each with two confidence axes (observed vs normative), a validation status, and PROV-style evidence. Code-observed relationships are never treated as ground truth: only oracle evidence (tests/CI/docs/human) raises normative confidence. Four lifecycle hooks retrieve validated/unverified/contradicted items before edits, update records without inventing facts, and consolidate on Stop; detects contradictions, proposes located fixes, and improves normative correctness over time. Every triple is validated against a predicate ontology (RDFS-style domain/range) before it enters the ledger — hallucinated verbs and semantically impossible pairings are rejected, not stored. Ships a 108-test install gate. |\n\nSee each skill directory's `SKILL.md` for usage and prerequisites.\n\n## Related projects\n\n[**Manifold**](https://github.com/dhanesh/manifold) — a constraint-first development\nframework that makes a feature's constraints explicit and machine-checkable *before* code\nis written, and keeps them checkable afterwards. Two skills here sit next to it:\n\n- [`spec-first-planning`](spec-first-planning/) does the same job inside one session — a\n  falsifiable spec, then tasks that each name their own verification. Manifold persists\n  that structure across sessions and enforces it with a CLI (`manifold validate`,\n  `manifold verify --verify-evidence`) and CI.\n- [`agent-ready-rails`](agent-ready-rails/) grades whether a repo gives agents a real\n  feedback loop. A Manifold-managed repo is one way to supply the R1/R2 rails it scores.\n",
  "bytes": 10924,
  "sha": "c26508f60c58e43a4c549c1d7208ed48880848a1e91080b41aff49ae014b0d54",
  "repo_slug": "dhanesh/agent-skills",
  "fonte": "repo",
  "truncated": false,
  "api": "https://agentalog.com/api/listings/okf_dhanesh_agent_skills_docs_knowledge_inde_bb26cf77/readme"
}