Compound V is a skill set for Claude Code. It covers the judgment around the code: what's worth building, how it should feel, how to keep a long session sharp, how to review for bugs and security without flattery, and how to tell whether an AI feature actually works.
The bet is that code got cheap and judgment didn't. So most of these skills are about the decisions around the typing, not the typing. They're short on purpose. If a line doesn't change what the agent does, it's cut, and the kit can run its own check (below) to keep it that way.
Every session starts at the router. using-compound-v loads up front and sizes the task first, so a typo just gets fixed and a real feature earns the full pipeline. A one-line change never spawns four agents.
For a real feature the path is short: pin the design with brainstorming, turn it into a plan an implementer with no prior context can follow with writing-plans, build it with batched-implementation (one fresh subagent per two or three related tasks, all on your strongest model), and review each batch with recheck before the next one starts. A five-task plan lands in about four dispatches.
using-compound-v → brainstorming → writing-plans → batched-implementation ⇄ recheck → finishing
(route the tier) (design gate) (plan or PRD) (1 impl / 2-3 tasks) (read-only) (merge/PR)
Two pieces carry most of the weight:
batched-implementationruns one implementer per two or three related tasks, all on your strongest model. It keeps going instead of stopping to ask permission, and reports each batch with a four-status contract.recheckis a single read-only pass, ordered cheapest-disqualifying-first: goals, plan, bugs, vulnerabilities, re-test, over-engineering. It returns severity-tagged findings and one verdict, and caps the fix loop at three rounds. It stays read-only because a reviewer that can edit ships its own unreviewed bug.
| Group | Skills |
|---|---|
| Foundation | using-compound-v: the router. Tiering, the taste/distribution/primitive gate, the non-negotiables. |
| Solve any goal (opt-in) | frame-the-goal (turn any goal into a testable success check) → simplest-thing-that-works (the simplest mechanism that provably passes it — zero-AI first, climb only when forced, as high as a hard goal needs) → make-it-stable (make it hold every time). The general front-door; caps the machinery, never the goal, and routes into the AI skills. |
| Taste | startup-taste (whether and what to build) and product-taste (how it feels) |
| Plan | brainstorming (design before code), writing-plans (a per-build plan with real code, no placeholders), and writing-prd (the product's stable source-of-truth doc, read first for context) |
| Thinking | critical-thinking (red-team your own reasoning before you commit — steelman it, hunt disconfirming evidence) |
| Build | batched-implementation, recheck, finishing |
| Correctness and security | test-driven-development, systematic-debugging, verification-before-completion, and agent-security (build-time defense: the lethal trifecta, source-trust, sandboxing model-written code) |
| AI design (one feature) | designing-agents (a call, a workflow, or an agent?), evals (does the AI actually work?), context-engineering |
| AI systems (architecture, opt-in) | architecting-ai-systems (the compound system around the model — harness-as-moat, primitive-not-wrapper, build for the model ~18 months out) and ai-system-reliability (keep a built system from corrupting its own state; chain a constellation past one model's ceiling) |
| Power | searching-patterns (the canonical pattern and the anti-pattern it replaces, from primary sources) and dispatching-parallel-agents |
Three rules sit above the skills:
- Honest. Evidence over claims, no praise-padding, no false "done." When something doesn't work, it says so.
- Safe. Security is a review axis that blocks a merge. It's never traded away to ship, and the kit won't write harmful code.
- Grounded. The skills come from how production coding agents actually behave and from primary engineering sources, not invented best practice. Every load-bearing number maps to a public source in
references/sources.md. A claim that can't be grounded is marked as a judgment call.
/plugin marketplace add LeventySeven/compound-v
/plugin install compound-v@compound-v-devFor local work, symlink the skill directories into ~/.claude/skills/. A SessionStart hook injects the small router each session and everything else loads on demand, so the always-on cost is the router alone.
The kit checks itself. bash scripts/check.sh reads every skill and fails if one breaks a rule: a body over its line budget, a frontmatter name that doesn't match its directory, a cross-reference to a skill that doesn't exist, an @path link, or any mention of the private research notes that must never ship. It has no dependencies, so it drops straight into CI or a pre-commit hook.
Compound V was built with its own loop: batched Opus implementers and a read-only recheck on every batch. The source material was an audit of how today's production coding and research agents behave, the canonical engineering and skill-authoring writing, practitioner talks, and a founder-judgment canon. The build's own recheck caught real defects, including a router over its line budget, a verdict-handling gap, and two descriptions that broke the kit's own rules, and they were fixed before commit. Each release re-checks discoverability and the review pass with a small eval suite.
MIT