What problem does it solve? Agents starting a session often miss the right documentation because context files are scattered across process/context/, development-protocols/, and feature folders. This Skill deterministically discovers and loads only the relevant context for the current task, avoiding both missed docs and wasteful whole-file reads. ## Core Features & Use Cases - Frontmatter-only discovery: The discover-context.mjs script lists all nested files under context, protocol, feature, and plan roots, extracting only YAML frontmatter (name, description, keywords, related) instead of reading entire files. - Keyword-first routing: The --match flag tokenizes a task description and ranks context docs by frontmatter keyword overlap, then appends related sibling docs for cross-domain tasks. - Drift-proof routing tables: The --emit-routing and --check-routing flags regenerate and lint the GENERATED routing block in all-context.md so the index never goes stale. - Use Case: At the start of a planning session for an authentication feature, run the script with --feature auth and --match "update the user ORM model" to get a ranked list of exactly the context docs, protocols, and plan files to read. ## Quick Start Ask the agent to run node .claude/skills/vc-context-discovery/scripts/discover-context.mjs with your feature name or task description to load the relevant context for this session.
