Curated skills, agents, and hooks for shipping software — codebase audits, plan review, build workflows, and infrastructure operations. Packaged for Claude Code (via the Claudfather marketplace), Cursor, and Codex. See the generated support table and lifecycle guide for each host's selected skills, prerequisites, update/removal paths and evidence limits. The local package doctor checks the loaded package without launching tools or changing settings.
Requires Claude Code (npm install -g @anthropic-ai/claude-code).
Inside Claude Code:
/plugin marketplace add Claudfather/clauDNA
/plugin install claudna@Claudfather
That's it. Skills become available as /claudna:<skill-name> (namespaced under the plugin). Tab completion works — type /claudna: and press Tab to browse, or /claudna:au and Tab to narrow. Hooks (auto-format, pretooluse permission expansion, notifications) activate automatically when the plugin is enabled.
The Cursor manifest is .cursor-plugin/plugin.json. It ships skills/ (minus a few host-/context-restricted skills, #340) and agents/, and deliberately no hooks. The shell hooks in plugin-hooks/ are wired through the respective Claude Code and Codex manifests; Cursor remains hook-free. Skills appear under Agent Decides and are invocable as /<skill-name> in chat.
Three install paths, depending on what you have:
- From the Cursor marketplace — open Customize in the sidebar, search for
clauDNA, and choose Install, picking a project or user scope. (Pending Cursor's review — see CONTRIBUTING § Submitting to the Cursor marketplace.) - As a team marketplace (Teams and Enterprise) — Dashboard → Plugins & MCPs → Add Marketplace → Import from Repo, with
https://github.com/Claudfather/clauDNA. Enable Auto Refresh so pushes tomainre-index the plugin. This path works today and does not wait on review. - From a local checkout — copy or symlink the repo into
~/.cursor/plugins/local/claudna, then run Developer: Reload Window. Cursor only follows symlinks whose target resolves inside that folder, so copy the tree if a symlink to your checkout is ignored.
Use the supported Codex plugin marketplace interface to add Claudfather/clauDNA
and install claudna@Claudfather. The manifest explicitly selects its workflow skills:
Claude-only cleanup and the two source-checkout contribution skills are excluded.
Use skill mentions or read the installed skill file; all selected entrypoints link
the Codex tool mapping. All eight role prompts map to a read-and-dispatch procedure, with no native agent
registration. A route is supported only where actual same-host controls and the
required fixture proof pass; unsupported routes remain explicit. A desktop result
does not establish CLI support.
Hooks require a separate operator trust action. This set preserves five store handlers and adds only the startup briefing; it carries no Claude permission, formatting or notification hooks. Native discovery, execution and stdout delivery are distinct checks. See Codex setup for trust, opt-outs and verification limits.
Optional, in any order:
- Recommended
settings.jsontweaks — permissions, statusLine, sandbox config. See SETUP_GUIDE §3. - Snowflake setup — only if you use the Snowflake skills. See SETUP_GUIDE §5.
- Shell aliases — power-user worktree commands. See SETUP_GUIDE §6.
- Project setup — in any project, run
/claudna:init-projectto generateCLAUDE.mdandCHANGELOG.md.
For bots, CI runners, and Docker images, drop a settings.json with enabledPlugins + extraKnownMarketplaces, then set CLAUDE_CODE_SYNC_PLUGIN_INSTALL=1 and run claude -p. Full recipe in SETUP_GUIDE §4.
| Directory | Count | Contents |
|---|---|---|
skills/ |
Derived by host | User-invocable slash commands |
agents/ |
8 | snowflake-analyst, dbt-engineer, neon-analyst, modal-ops, railway-ops, vercel-ops, code-reviewer, spec-reviewer |
plugin-hooks/ |
6 wired + 1 opt-in | SessionStart briefing (opt-out CLAUDNA_SESSION_BRIEFING=0), auto-format on Write/Edit, PreToolUse permission expansion, PreCompact capture gate, macOS notifications, and the session store (boundaries, activity, summaries, draft harvest and local skill telemetry, all on by default; SETUP_GUIDE §3.7). (statusline.sh is opt-in — see SETUP_GUIDE §3.2.) Named plugin-hooks/ to avoid a Claude Code bug that deletes any project-root hooks/ directory between tool calls. |
Codex receives its manifest-selected skill set and store/startup handlers; role prompts are read-and-dispatch only. Claude Code gets all three component types. Cursor gets skills/ (minus a few host-/context-restricted skills, #340) and agents/ only — see Cursor above. The plugin-hooks/ name does double duty: besides dodging the Claude Code bug, it keeps the hooks out of hooks/, which is where Cursor's folder discovery would otherwise find and wire them.
Skills are thinking frameworks, not SKUs. Before adding a new skill, ask: can this be a lens or mode inside an existing skill? Every skill description loads into every session, so fragmenting one capability into per-action skills bloats context and hides the framework the agent should be thinking in. Prefer one skill with modes over N near-duplicates — /adversarial-review runs single, dispatch, and respond modes; /ironclad fields an entire review panel through one entry point.
Descriptions are routing surfaces. The description field is what the model reads when choosing a skill, so it states when to reach for the skill, never how the skill works internally: trigger conditions first, confusable siblings disambiguated in place ("For X, use /claudna:Y"), CLI flags documented only in argument-hint. The grammar is contract-bound and validator-enforced — see SKILL_CONTRACT.md §2.1.
Invocable as /claudna:<name> after marketplace install (e.g. /claudna:audit tech-debt). Bare names below are shown without the prefix for readability.
On a host without Claude Code's namespaced commands (Cursor, or an agent that reads skills/ directly), read /claudna:<name> in skill text as the skill <name>, invoked however that host invokes skills (SKILL_CONTRACT §2.1).
| Skill | Description |
|---|---|
/session |
Session-continuity engine — resume / handoff / checkpoint / name modes (replaces the three session-*/name-session skills) |
/build |
Execute a design doc or GitHub Issue — browse, pick, challenge, build, test, PR |
/build-all |
Drive an epic's worth of ordered phase docs as one governed multi-PR run |
/worktree |
Manage git worktrees for parallel Claude sessions |
/delegate |
Bounded task between Claude Code and Codex in an owned workspace; existing parent/child history and automatic bookkeeping cleanup |
/sync-branches |
Fetch and merge origin/main into every branch and worktree across sibling repos |
/development-retro |
Post-merge retrospective — systemic patterns, friction, breadcrumb trails |
/weigh-development-paths |
Compare multiple viable approaches at a development junction |
/adversarial-review |
Structured plan challenge with evidence and anti-groupthink guards |
/forge |
Plan a multi-phase workstream — decision forks, phasing, spec/roadmap authoring |
/ironclad |
Harden a plan Issue or PR with a panel of independent review lenses — seven panel lenses built in; --lens <name> for a single-lens report |
| Skill | Description |
|---|---|
/audit |
Codebase-audit engine — security / tech-debt / booby-traps / docs / design / access-path / data-model / data-model-redesign / frontend-perf / repo-health / system / cache / scale lenses (replaces the eight standalone audit skills) |
/review-work |
Review engine — changes (uncommitted work) / pr (one PR) / multi-pr (parallel subagent reviews + synthesis) modes (replaces the two review skills) |
/verify-completion |
Verify work is actually complete before claiming success |
/qa |
Headless-browser QA — test-and-fix, report-only, canary, benchmark, browse, cookie-setup, or autonomous visual crawl |
| Skill | Description |
|---|---|
/kindle |
Ideation gate before any creative work — discover intent into an approved spec, hand off to /forge |
/product-enhance |
Discover enhancements via intent interviews and gap analysis |
/product-vision |
Architecture-aware product vision — explore what the codebase could become |
| Skill | Description |
|---|---|
/capture |
Save knowledge to the vault — a note you write, external content (URL / file / text), or the current session's learnings (bare /capture) |
/index |
Organize the shared knowledge store; regenerate INDEX.md |
/recall |
Orientation briefing — what the fleet and this project already know, before you start |
/publish |
Distribute knowledge to other surfaces (bots, docs sites) |
/claudron |
Claudron vault engine — lookup / status / doctor over the shared knowledge vault (needs the Claudron CLI); doctor applies the engine's migrations only after you confirm; save via /capture |
| Skill | Description |
|---|---|
/ship |
Commit, push, and open a PR (ship commit for a local-only checkpoint) |
| Skill | Description |
|---|---|
/dbt |
dbt build, test, compile, full-refresh |
/neon |
Neon PostgreSQL engine — query / branch / info modes (replaces the three neon-* skills) |
| Skill | Description |
|---|---|
/modal |
Modal engine — deploy / logs / status modes (replaces the three modal-* skills) |
/railway |
Railway engine — deploy / logs / status modes (replaces the three railway-* skills) |
/vercel |
Vercel engine — deploy / logs / status modes (replaces the three vercel-* skills) |
/investigate-app |
Platform-agnostic production debugging |
| Skill | Description |
|---|---|
/find-skills |
Discover and install new agent skills |
/heist |
Raid a GitHub repo for skills, config patterns, and novel approaches |
/init-project |
Set up CLAUDE.md, CHANGELOG.md, and .claude/ for a new or existing project |
/file-github-issue |
File a GitHub issue from a screenshot + short description |
/github-activity-report |
GitHub activity stats (PRs, commits, contributors) over a time window |
/using-claudna |
Orient in the skill set + Installation Health diagnostic (replaces the old skill-health diagnostic) |
/promotion-intake |
Land a Claudosseum promotion package as a reviewable PR (validator-gated) |
/skill-scaffold |
Scaffold a new clauDNA skill directory with valid frontmatter |
/cleanup-legacy-install |
Remove a legacy pre-plugin ~/.claude/ overlay after upgrading |
The skills form a pipeline:
Morning
───────────────────────────────────────────────────
/claudna:audit repo-health → See what needs attention across all repos
/claudna:session resume → Pick a repo, get a 30-second briefing
Planning (when starting new work)
───────────────────────────────────────────────────
/claudna:audit tech-debt → Find what's broken → generates phase docs
/claudna:product-enhance → Find what could be better → generates phase docs
/claudna:product-vision → Explore what the codebase could become
Building (the core loop)
───────────────────────────────────────────────────
/claudna:build → Point at a phase doc
├── Challenge round (first principles review)
├── Build & test
├── PR with deliverable audit
└── Say "next" → auto-compacts → next phase
Reviewing
───────────────────────────────────────────────────
/claudna:review-work pr 42 → Structured review of any PR
/claudna:review-work changes → Audit uncommitted work
End of Day
───────────────────────────────────────────────────
/claudna:session handoff → Capture decisions, learnings, next steps
→ Writes handoff file for tomorrow's /claudna:session resume
First time? Start with /claudna:audit repo-health to see your repos, then /claudna:session resume in any project.
Specialized personas invoked by skills or directly:
"Use snowflake-analyst to find the largest tables"
"Use dbt-engineer to review this model"
"Use neon-analyst to check table sizes"
"Use railway-ops to diagnose the deployment failure"
"Use vercel-ops to investigate the 500 errors"
"Use modal-ops to check GPU utilization"
"Use code-reviewer to assess this PR for correctness and test coverage"
"Use spec-reviewer to verify the implementation matches the plan"
The marketplace install is managed by Claude Code: plugins live under ~/.claude/plugins/cache/Claudfather/claudna/<version>/. Updates land in a new version directory; old versions are kept for ~7 days before auto-cleanup.
Never touched by claudna:
~/.claude/settings.json— user-managed. claudna never writes here. Recommended snippets are documented in SETUP_GUIDE for you to merge manually.~/.claude/notes/— your personal files. claudna never writes here; the session engine's resume mode can copy a legacynotes/projects/<slug>/context-resume.mdhandoff into a project with no handoff yet. It preserves the source unchanged, and subsequent resumes use the project copy.
project-template/ provides starter project instructions and .claude/ settings. These are packaged setup resources; enabling the plugin does not apply them to your project. Copy them manually or use /claudna:init-project:
| File | Purpose |
|---|---|
CLAUDE.md |
Project instructions with workflow principles |
.claude/settings.json |
Project-specific permissions |
See CONTRIBUTING.md for the full guide. Writing a new skill? Start with the Skill Authoring Guide.
Quick version: branch off main, make your change, update CHANGELOG.md, run make check (the exact check-set CI runs), open a PR.
Working in a Cursor Cloud Agent? .cursor/environment.json installs the pinned check toolchain on boot, so make check runs without setup — see CONTRIBUTING § Cloud Agents.
| File | Description |
|---|---|
| PROJECT_MISSION.md | What clauDNA is, what it's becoming, and where it fits in the Claudfather ecosystem |
| CLAUDE.md | Operating rules for agents extending clauDNA |
| SKILL_CONTRACT.md | The binding contract every skill must satisfy. Enforced in CI by scripts/validate-skills.py |
| AGENT_CONTRACT.md | The binding contract every agent must satisfy. Enforced in CI by scripts/validate-agents.py |
| SETUP_GUIDE.md | Deep-dive setup — settings.json bootstrap, headless/CI provisioning, Snowflake key-pair auth, troubleshooting |
| SKILL_AUTHORING_GUIDE.md | How to write, test, and submit a skill — patterns, examples, anti-patterns |
| CONTRIBUTING.md | Contribution workflow — bug reports, PRs, testing requirements |
| CHANGELOG.md | All notable changes |
Licensed under the Apache License, Version 2.0. See NOTICE for the copyright notice that travels with it.