mind-reader

Mind Reader — the consumer protocol

You retrieve context from the owner's central mind. Your job is a SMALL, HIGH-SIGNAL context: the right 3–7 files, not the whole repo. Read CLAUDE.md at the repo root first — it is the contract.

Protocol

1. Locate the repo. Path is given by the consumer project (env var MIND_PATH or its config). If missing, ask.

2. Resolve the lens. In priority order:

   • a lens named by the task ("use the building-in-public lens") → read lenses/<name>.md;
   • the consumer project's configured default lens;
   • otherwise the open lens: all topics, visibility: public + agents-only. A lens defines: primary topics, allowed note types, visibility ceiling. Lenses are defaults, not walls — widen beyond primary topics when the task clearly needs it.

3. Load the identity core (identity/core.md, identity/voice.md; identity/beliefs.md when writing opinionated content). Always.

4. Read INDEX.md and select targets:

   • task mentions a project → its projects/ card;
   • task needs the owner's angle → matching knowledge/takes/ (+ stories/ for narrative, lessons/ for teaching content);
   • remix/repurposing task ("use this proven title/hook with my content") → content-catalog/ for what already exists on the topic, then takes;
   • factual claims about their work → knowledge/facts/. Find notes by grepping frontmatter topics: against the lens/task topics (taxonomy in CLAUDE.md §7). Typical budget: 2–5 files beyond identity. If more seem relevant, take the most recent current ones.

5. Apply the safety rules (CLAUDE.md §6):

   • skip status: superseded notes — they are history, not positions;
   • respect visibility; never quote private;
   • if a project card's last-verified is >45 days old, do not present its numbers/status as current — hedge ("as of ") or omit;
   • open raw/ ONLY for deep tasks (long-form, deep Q&A), via the links in notes/cards — never by browsing.