brain-protocol

Load and maintain Snowshoe's spec-driven brain. Use at session start, before any behavior or product-claim change, and when CURRENT/ADRs may be stale.

SKILL.md

GitHub

Brain protocol

Read order (required)

  1. docs/brain/START_HERE.md
  2. docs/brain/CURRENT.md — compiled truth
  3. docs/brain/ROADMAP.md — sequence of intent, not shipped fact
  4. docs/brain/EXPERIMENTS.md — registry of bets, not truth

Then open only the ADRs, notes, or evidence the task needs.

Trust

  • Truth = CURRENT.md + ADRs with status: accepted (active is a lint alias only; DECISIONS table uses accepted).
  • Experiments stay in the registry until promoted.
  • docs/evidence/** supports a claim; it does not become true by existing.
  • docs/brain/notes/** is working memory.
  • docs/archive/** is historical.

Writes (required on behavior / claim change)

If you change product behavior, positioning, non-goals, monetization, or any claim CURRENT makes:

  1. Add or update an ADR under docs/brain/DECISIONS/ (use docs/brain/templates/DECISION_TEMPLATE.md).
  2. Update CURRENT.md in the same change so it stays the compiled truth.
  3. Set frontmatter status: correctly (accepted, proposed, superseded, …).
  4. Move superseded canon to docs/archive/ rather than deleting history silently.

Do not implement application code until CURRENT says we are past scaffold-only.

Hygiene

  • Keep CURRENT short. Detail lives in ADRs.
  • Do not invent metrics, users, or competitors.
  • English in brain files.