Help us improve
Share bugs, ideas, or general feedback.
Share bugs, ideas, or general feedback.
Share bugs, ideas, or general feedback.
By subagentceo
Per-merged-PR audit follow-up generator. Reads the session transcript at ~/.claude/projects/<dir>/<session-id>.jsonl, computes tool-call waste/latency/throughput stats for the work that landed a given PR, and emits a templated follow-up PR body plus the top 3 highest-impact findings as issue bodies. Dogfoods OAUDIT family: build-use-iterate-improve the merge process itself by surfacing where the agent wasted calls.
npx claudepluginhub subagentceo/knowledge-engineering --plugin pr-audit-templateGiven a ranked-findings list from waste-analyze, render a standardized PR follow-up body + N issue bodies (default 3) using a handlebars-style template. Each issue is actionable — names the wasteful pattern, quantifies impact, and suggests a concrete fix or workflow change.
Read a Claude Code session JSONL transcript and emit a structured list of tool calls with start/end timestamps, inputs (redacted), and duration. Use when computing throughput metrics for a merged PR — call this first, then pass its output to waste-analyze.
Given a structured tool-call list from transcript-parse, identify the top sources of wasted calls — re-reads of the same file, retried commands that failed repeatedly, subagent dispatches that returned nothing useful, monitor loops that fired without acting on output. Output a ranked list of findings with impact scores.
This plugin requires configuration values that are prompted when the plugin is enabled. Sensitive values are stored in your system keychain.
gh_repoWhere the audit PR + issues land. Defaults to this chassis.
${user_config.gh_repo}Share bugs, ideas, or general feedback.
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge.
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge.
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
top_findingsHow many highest-impact findings to surface as follow-up issues per merged PR. Operator convention is 3.
${user_config.top_findings}transcript_rootPer-host directory containing per-session JSONL transcripts. Defaults to ~/.claude/projects/<project-slug>/.
${user_config.transcript_root}UI/UX design intelligence. 67 styles, 161 palettes, 57 font pairings, 25 charts, 15 stacks (React, Next.js, Vue, Svelte, Astro, SwiftUI, React Native, Flutter, Tailwind, shadcn/ui, Nuxt, Jetpack Compose). Actions: plan, build, create, design, implement, review, fix, improve, optimize, enhance, refactor, check UI/UX code. Projects: website, landing page, dashboard, admin panel, e-commerce, SaaS, portfolio, blog, mobile app. Elements: button, modal, navbar, sidebar, card, table, form, chart. Styles: glassmorphism, claymorphism, minimalism, brutalism, neumorphism, bento grid, dark mode, responsive, skeuomorphism, flat design. Topics: color palette, accessibility, animation, layout, typography, font pairing, spacing, hover, shadow, gradient.
Comprehensive skill pack with 66 specialized skills for full-stack developers: 12 language experts (Python, TypeScript, Go, Rust, C++, Swift, Kotlin, C#, PHP, Java, SQL, JavaScript), 10 backend frameworks, 6 frontend/mobile, plus infrastructure, DevOps, security, and testing. Features progressive disclosure architecture for 50% faster loading.
This skill should be used when users need to generate ideas, explore creative solutions, or systematically brainstorm approaches to problems. Use when users request help with ideation, content planning, product features, marketing campaigns, strategic planning, creative writing, or any task requiring structured idea generation. The skill provides 30+ research-validated prompt patterns across 14 categories with exact templates, success metrics, and domain-specific applications.
Develop, test, build, and deploy Godot 4.x games with Claude Code. Includes GdUnit4 testing, web/desktop exports, CI/CD pipelines, and deployment to Vercel/GitHub Pages/itch.io.
Upstash Context7 MCP server for up-to-date documentation lookup. Pull version-specific documentation and code examples directly from source repositories into your LLM context.
Comprehensive startup business analysis with market sizing (TAM/SAM/SOM), financial modeling, team planning, and strategic research
Code-intelligence + document tooling for mirroring HTML/PDF documentation (e.g. anthropic.com/sitemap.xml pages): HTML, TypeScript, and JavaScript language servers (VS Code's extracted servers + typescript-language-server) plus the official Model Context Protocol PDF server. Binaries are pinned as repo devDependencies and invoked via npx.
Operator-side IT admin plugin for GitHub-tier operations: claude-code OAuth rotation, repo→org secret promotion, secrets parity audit, branch-protection ruleset CRUD, and a security-aware linter for Claude Code Action workflows (.github/workflows/claude*.yml). Dogfoods OSEC1/OSEC2/OSEC3 (secret model), OIT1 (minter pattern), and OAUTO13 (workflow security baseline) into a single plugin that exercises the full claude-code plugin spec: skills + hooks + monitors + agents + MCP server.
Personalized platform-engineering plugin for the operator: AlloyDB Omni install, Docker MCP profile wiring, Voyage+Turbopuffer+AlloyDB embeddings architecture, and citation/test/outcome discipline. Modeled after subagentceo/knowledge-work-plugins/cowork-plugin-management.
Operator-side IT admin plugin: scriptable CRUD over vendor API tokens for Cloudflare, Turbopuffer, Neon, Parallel.ai, and Nimbleway. Implements the OSEC3 long-lived-minter pattern with macOS keychain bootstrap. Replaces browser-click token mint flows with deterministic CLI flows that never leak secret values into agent context.
Solo-founder chassis for shipping a Claude-powered product. Multi-agent research orchestrator + 28 vendor doc mirrors + 16+ MCP tools across 5 lanes + Cloudflare Sandbox runner + Neon-branched per-PR previews. OAuth-only.
This repo is a fork-and-ship chassis, not a one-off project. The intent (per PRODUCTRD.md) is that another founder clones the repo, swaps the seed prompts and vendor list, and inherits everything else: the verify chain, the heartbeat memory layer, the auto-merge loop, the citation discipline, the operator runbooks, and the OAuth-only posture.
| Surface | What | Where |
|---|---|---|
| Orchestrator | Opus 4.7 (1M context) — 4 sub-agents over @anthropic-ai/claude-agent-sdk | src/agent/run.ts |
| MCP tools | 16+ tools across 5 lanes: engineering_*, blog_*, support_*, llms_*, vendor_* + search_tools | src/mcp/ |
| Vendor mirror | 28 vendor doc surfaces (anthropics, cloudflare, neon, stripe, twilio, workos, elevenlabs, aws, openfeature, gcp, ...) — 1,369 anthropics docs alone | vendor/ |
| Crawler | crawlee + llms.txt / html-index / sitemap.xml discovery; preflight-304 idempotency | scripts/crawl-vendors.ts |
| Worker runner | Cloudflare Sandbox + Durable Objects for per-task ephemeral execution (scaffolded) | infra/cloudflare/ |
| Neon branching | Per-PR Neon DB branches via cloudflare-preview.yml | migrations/, scripts/migrate-neon.ts |
| Frontend | outcomesdk.com Cloudflare Worker — pretext-driven SPA over vendor/ markdown | frontend/ |
| Heartbeat memory | Cross-session orchestration state | seeds/memory/heartbeat/ |
| Feature flags | OpenFeature + Cloudflare Flagship provider | src/lib/openfeature.ts |
| Plugin manifest | 3 Anthropic marketplaces (official, knowledge-work, community) | .claude/plugins.json |
unset ANTHROPIC_API_KEY # OAuth-only — fails closed if this is set
export CLAUDE_CODE_OAUTH_TOKEN=... # mint via `claude setup-token`
npm install
npm run verify # mcp + tf + citations + gates + libs + freshness + project
npm run dev "trivial test query" # local orchestrator turn
See DEVELOPER.md for the full first-time setup + day-to-day workflows.
| Doc | When to read |
|---|---|
CLAUDE.md | A Claude session starting in this repo — load-bearing context auto-loaded by claude |
DEVELOPER.md | First-time setup; adding a vendor / lane / skill / test |
RUNBOOK.md | Using Claude Opus 4.7 1M context as the web orchestrator |
CONTRIBUTING.md | Forking-founder onboarding + PR discipline |
docs/architecture.md | Runtime topology |
docs/governance.md | Branch ruleset + auto-merge state machine |
docs/security.md | OSV-Scanner dependency-vuln gate posture |
docs/context-management.md | Token counting, cache boundary, settingSources, safety hooks |
docs/CONVENTIONS.md | Outcome-driven Conventional Commits |
docs/PROJECT.md | Cowork-style project manifest |
docs/pending.md | Live action dashboard — operator + agent queue |
docs/operator-runbooks/README.md | Claude-in-Chrome operator runbooks (CF API token, GH PAT, etc.) |
PRODUCTRD.md | Chassis intent + functional requirements |
SUBPROCESSORS.md | Vendor inventory for fork-time re-evaluation |
| Lane | Source | Tools |
|---|---|---|
engineering | anthropic.com/engineering | engineering_{index,fetch,search} |
blog | claude.com/blog | blog_{index,fetch,search} |
support | support.claude.com | support_{collections,collection,article} |
llms | namespaces under *.claude.com/llms.txt, anthropic.com/llms.txt, vendor llms.txts | llms_{namespaces,fetch,grep} |
vendor | the local vendor/ mirror (28 surfaces) | vendor_{list,fetch,grep} |
Plus search_tools for progressive disclosure across the surfaces.
The full lane-to-tool map is in docs/architecture.md. Per-lane docs at docs/lanes/{engineering,blog,support,llms,vendor}/index.md.
| Sub-agent | Tools | Purpose |
|---|---|---|
npm-research | 4 npm-registry MCP tools | Primary npm data; cites registry URLs |
verifier | 12 knowledge-bridge tools (excl. vendor_*) | Independent grader vs docs/rubric.md |
crawl-curator | 3 vendor_* tools | Per-vendor crawl.json audits + drift detection |