π
Plugins
Loading...
Showing 30 of 23040 results
F1 strategy engineer for your LLM garage: analyzes a repo, reads your hardware, and recommends the optimal local LLM models.
Coding harness for long-term tasks, based on @claude article harness-design-long-running-apps
Planner / Generator / Evaluator loop for harness-driven development with split specs, sprint contracts, and per-sprint progress.
Three-role harness β planner sets outer lines, generator builds, evaluator operates the artifact and grades against a weighted rubric. Each role runs as a fresh subagent with its own context window.
Harness β a lightweight reasoning floor. Pass a raw request; a fixed six-stage engine runs Plan(opus) β SetGoal(opus, adversarial critic) β Implement(sonnet, repo-skill executors) β Test(sonnet, deterministic evidence) β QualityGate(opus, bounded loop + goal-level gate) β Report(sonnet). Filters repetition and below-threshold answers regardless of main-session model. B (request + engine) default; A (bespoke Workflow) for custom control flow. Enforcement: opt-in PreToolUse gate β projects list gated paths in .claude/harness-gate.json; edits there require harness engagement (fail-open). The install skill scaffolds a project: it embeds the self-contained gate hook (copies goal-gate.mjs into .claude/hooks/ and merges a PreToolUse entry into committed .claude/settings.json), plus gate config, default conventions (coding/verification/boundaries), and an ambient CLAUDE.md section β project-owned copies, never overwritten. It also optionally embeds the engine + statically-referenced skills into .claude/harness/ for plugin-less environments (opt-in; dynamic subgoal skills[] can't be pre-enumerated). Mode M: for control flow the fixed stages can't express, the harness generates a bespoke Workflow from templates/meta-skeleton.js (contract preserved: judgeβ actor, model pins, bounded loops, deterministic Test, goal-level gate).
Read-only advisor agents (architect, designer, reviewer, perf, principles, docs), profile-driven /pr and /ticket skills, enforcement hooks, and /harness-init.
Claude Code νλ€μ€ μ§ν λꡬ: CLAUDE.md(κ·μΉ μ λ¬) + Permissions(μν μ°¨λ¨) + Hooks(μλ κ²μ¦) + MCP(ν μ€νΈ λꡬ) + Subagent(AI λΆλ¦¬) + μ§ν μμΉ(λ©ν) 6λ¨κ³λ₯Ό μ§λ¨Β·μΆμ²Β·νμ μν¬νλ‘μ°λ‘ λ¨κ³μ μΌλ‘ 보κ°ν©λλ€. μ κ· μ μ 보λ€λ μ΄λ―Έ μ°λ νλ‘μ νΈμ νλ€μ€λ₯Ό μ§νμν€λ λ° μ΄μ . 1κ° μ€ν¬.
Warn-only, zero-dependency runtime harness layer for Claude Code (companion to superpowers): persistent state, scope control (prompt + action side), Default-FAIL verification & coverage gates, entropy governance (golden-rules + drift detection), and first-class C/C++ engineering (CMake/Meson/Make/Bazel + clang-tidy + sanitizers).
Four non-negotiable safety hooks plus cross-cutting engineering modules as auto-loading skills. Every domain pack depends on this.
v1.0.1 | κ°λ° μ€κ°μ μμ΄μ νΈ μ μ΄ μΈνλΌ(ν , verifier, λ¬Έμ μλν) μνλ₯Ό λ 립 μλΈμμ΄μ νΈκ° μ§λ¨νκ³ , λ¬Έμ λ°κ²¬ μ upstream μμ λ ν¬νΈλ₯Ό μμ±νλ νλ€μ€ μ κ² μ€ν¬
Three defense-in-depth hooks: credential value scrub (PostToolUse), admission keyword reminder (UserPromptSubmit), bash command guard (PreToolUse). Built after repeated credential leak incidents β structural fix when behavioral rules in CLAUDE.md keep failing.
Curated data pack: analytics engineering, analyst notebook, LLM app, ML pipeline.
Authoring aids for building Claude Code harnesses β loops, flows, and the specs that separate them.
Curated devops pack: CI/CD platform, infrastructure, Kubernetes platform, observability/SRE.