Context Mill
https://github.com/PostHog/context-mill- Category
- AI Agents
- Rank
- No. 675Tools index
- Pricing
- Open Source
- Type
- TOOL
- Builder
- PostHog
- GitHub
- 59 stars
- Latest release
- v1.51.0
- Date
About
PostHog's context and skill assembly toolkit for AI agents and LLM tasks. Builds focused context bundles for coding assistants.
What it does
Context Mill runs a publishing pipeline for PostHog knowledge. It collects raw documentation, maintained instructions, agent prompts, and simplified example applications, then turns them into self-contained skill archives. Builds also create manifests, command menus, agent prompt outputs, marketplace plugins, and a combined archive. Consumers fetch individual packages on demand through the PostHog wizard or an MCP-compatible client.
Why it's ranked here
This is credible infrastructure for teams maintaining a large, changing body of agent guidance. Automatic ordering, versioned release artifacts, command metadata, local incremental rebuilding, ownership routing, and pre-release threat scanning address real operational problems. Its value is narrower than a general context framework, however, because the supplied pipeline and content are centered on PostHog.
What's good
Each skill ships with its instructions, ordered reference steps, and shared documentation captured during the build. The manifest gives consumers stable metadata and download locations, while the menu maps skills into command families. Local development rebuilds only affected skills and avoids repeated documentation fetches. Built packages are scanned for prompt injection and other risky text, with optional language-model triage to separate threats from false positives.
Tradeoffs
The repository serves PostHog's ecosystem rather than offering a neutral knowledge ingestion product. Example applications deliberately use fake authentication and are explicitly unsuitable as starter projects or production code. Incremental development rebuilds do not refresh the combined archive or marketplace output until a full manual build. Security triage needs credentials; without them, every scanner match is treated as a threat.
How to use it well
It fits documentation and developer-experience teams that maintain PostHog instructions for agents, command-line workflows, or MCP clients. Keep skill steps small, ordered, owned, and reviewed, then publish generated archives through versioned releases. Use the example applications as teaching context for integration decisions. Do not treat Context Mill as a production application scaffold, authentication example, general-purpose crawler, or standalone MCP server.
Technical notes+
package.json defines a private Node.js ES-module package requiring Node 20.11 or newer, with build, development, test, visualization, and security-scan scripts. scripts/build.js generates per-skill ZIPs, bundled JSON groups, manifests, command menus, marketplace plugins, agent prompts, release-asset documentation, and dist/skills-mcp-resources.zip. scripts/dev-server.js watches context/skills/, context/agents/, and example-apps/, routes changes to incremental rebuilds, and serves ZIP, JSON, and Markdown artifacts over HTTP. scripts/plugins/index.js composes ordered content transforms and preserves input when a plugin throws. scripts/scan-warlock.js extracts built archives, scans supported text formats with Warlock, and optionally triages matches through an Anthropic client pointed at the PostHog gateway.
Observed
- Primary implementation language
- JavaScript using Node.js ES modules
- Runtime support
- Node.js 20.11 or newer
- Packaging surface
- Private npm package with pnpm configuration and ZIP, JSON, and Markdown build artifacts
- Interfaces
- npm CLI scripts, local HTTP development server, Agent Skills packages, and manifests consumed by MCP-compatible clients
- Testing
- Vitest suites cover plugin and skill-library code
- Example platforms
- Examples include React, Next.js, Django, React Native, Expo, Astro, Vue, Angular, and TanStack variants
Read from README.md, package.json, scripts/build.js, scripts/dev-server.js, scripts/visual-dags.js, scripts/scan-warlock.js, scripts/plugins/index.js, example-apps/react-native/App.tsx, example-apps/react-native/index.js, example-apps/astro-ssr/astro.config.mjs, example-apps/astro-hybrid/astro.config.mjs, example-apps/react-react-router-7-data/index.tsx, example-apps/vue-3/src/main.ts, example-apps/expo/app/index.tsx, example-apps/angular/src/main.ts.
What it can do
Build context bundles for AI coding assistants
Code repositories and project files → Focused context bundles
Assemble skills for LLM tasks
Task requirements and available skills → Skill assembly configurations
Extract relevant code context
Codebase and query parameters → Contextual code snippets and documentation
Filter and prioritize context information
Raw project context and relevance criteria → Prioritized context data
Generate context for AI agent workflows
Agent task definitions and data sources → Structured context packages
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.