By mbeacom
Local-first semantic and opt-in hybrid search: a bin/rag CLI that embeds through a configurable Ollama endpoint (default localhost), indexes with turbovec, and can fuse FTS5/BM25 candidates. Notes-first, corpus-agnostic via pluggable loaders.
Executables (bin/) — files in this plugin's bin/directory are added to the Bash tool's PATH while the plugin is enabled.
This plugin requires configuration values that are prompted when the plugin is enabled. Sensitive values are stored in your system keychain.
embed_modelollama model used to embed chunks and queries
${user_config.embed_model}ollama_hostBase URL of the ollama server
${user_config.ollama_host}Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
A context-engineering plugin pack for GitHub Copilot CLI, Microsoft's APM (Agent Package Manager), and Claude Code — getting the right information in front of an agent and keeping the wrong information out. It bundles complementary retrieval modalities — lexical, structural, code-intelligence, structured-data, history, semantic (RAG), graph, and durable memory — plus a routing agent that picks and composes them. Default workflows keep indexes and reviewed records on your machine. Configured model endpoints, providers, and allowlisted commands can still reach external systems. Around that spine it adds plan-execute (a strong model plans; cheaper subagents execute), context-steering (put each rule at the cheapest layer that still fires), verify (read-only claims and change-impact analysis), runtime-evidence (controlled escalation when static checks cannot settle a runtime claim), context-handoff (bounded cross-session task state), memory (provenance-bound durable recall with an optional MemPalace provider), and plugin-forge (author and quality-check portable plugins). The marketplace ships eleven plugins.
📖 Documentation site — install guides, architecture, and a page for every plugin.
Start with the task-oriented cookbook, review security and trust boundaries, or diagnose first-run and lifecycle problems in troubleshooting. Vulnerability reports follow the root security policy.
All three hosts install the same plugins directly from one marketplace — GitHub
Copilot CLI via copilot plugin, APM via apm install, and Claude Code via
/plugin — no manual copying of skill folders. The catalog ships in Claude Code's
marketplace schema, which Copilot and APM read too.
GitHub Copilot CLI installs these plugins directly from the marketplace — the same
OWNER/REPO this repo publishes:
copilot plugin marketplace add mbeacom/context-kit
copilot plugin install code-search@context-kit # auto-installs retrieval-core
copilot plugin install local-rag@context-kit
copilot plugin install obsidian@context-kit
copilot plugin install plan-execute@context-kit # plan-big/execute-small orchestration
copilot plugin install context-steering@context-kit
copilot plugin install verify@context-kit # auto-installs retrieval-core
copilot plugin install runtime-evidence@context-kit # pulls verify, then retrieval-core
copilot plugin install context-handoff@context-kit # pulls verify, then retrieval-core
copilot plugin install memory@context-kit # pulls handoff, verify, retrieval-core
copilot plugin install plugin-forge@context-kit
See docs/GITHUB_COPILOT.md for details, including the
local-rag CLI bootstrap outside Claude Code (Copilot does not run the plugin's
SessionStart hook).
APM installs the same plugins from the same marketplace, adding a committed lockfile, audit/policy checks, transitive dependency resolution, and cross-harness deploy. Register the marketplace, then install:
apm marketplace add mbeacom/context-kit
apm install code-search@context-kit # also pulls retrieval-core (the spine)
apm install local-rag@context-kit
apm install obsidian@context-kit
apm install plan-execute@context-kit
apm install context-steering@context-kit
apm install verify@context-kit # also pulls retrieval-core
apm install runtime-evidence@context-kit # pulls verify, then retrieval-core
apm install context-handoff@context-kit # pulls verify, then retrieval-core
apm install memory@context-kit # pulls handoff, verify, retrieval-core
apm install plugin-forge@context-kit
APM reads the repo's .claude-plugin/marketplace.json and each plugin's native
layout directly; each per-plugin apm.yml carries APM metadata and dependencies.
runtime-evidence and context-handoff depend on verify, which transitively
pulls retrieval-core; memory depends on context-handoff. See
docs/APM.md for
targets, the local-rag bootstrap (APM does not run Claude's SessionStart hook),
and maintainer notes.
/plugin marketplace add mbeacom/context-kit
Then install what you need (installing code-search auto-installs retrieval-core):
npx claudepluginhub mbeacom/context-kit --plugin local-ragProvenance-bound durable memory across sessions: immutable records, append-only review state, active-only recall, explicit MemPalace reconciliation, and opt-in lifecycle capture.
Controlled dynamic evidence collection for runtime claims that static verification cannot settle. Runs only pre-reviewed command IDs from a user-owned allowlist and emits bounded, verdict-ready artifacts.
Choose where guidance lives — always-on memory, path-scoped rules, on-demand skills, delegated subagents, MCP servers, or deterministic hooks — to steer agents while keeping the always-on context budget small.
Portable, bounded, verified task-state handoffs across agent sessions and hosts, with manual write/resume authority and an explicit validated boundary for optional historical memory archival.
Fast CLI search across modalities: lexical (rg/fd), structural (ast-grep/semgrep), code-intelligence (LSP/global/ctags), structured-data (jq/yq/gron), history (git pickaxe), structured rewrite (comby), data files (duckdb), metrics (tokei/scc), and non-code docs (rga/pandoc/pdftotext).
Ultra-compressed communication mode. Cuts 65% of output tokens (measured) while keeping full technical accuracy by speaking like a caveman.
Memory compression system for Claude Code - persist context across sessions
Multi-model consensus engine integrating OpenAI Codex CLI, Gemini CLI, and Claude CLI for collaborative code review and problem-solving.
45% cost reduction measured. Cache expiry prevention, SubTask auto-delegation, zero-cost context restoration, real-time cost dashboard. The only Claude Code plugin built from CC source analysis.