Meta Heartbeat

Heartbeat — Session Startup

Execute the following steps in order. Do not skip steps. Do not ask permission.

Step 1: Load Identity

1. Read context/SOUL.md — who you are, how you behave
2. Read context/USER.md — who you're helping and their preferences

Step 2: Load Memory

3. Read context/MEMORY.md — long-term curated knowledge (main sessions only)
4. Read today's daily file: context/memory/{YYYY-MM-DD}.md
5. Read yesterday's daily file: context/memory/{yesterday's date}.md

If either daily file doesn't exist, that's fine — skip it silently.

Step 3: Load Workflow Config

6. Read workflow.json if it exists — note the docs repo name and project type
7. If workflow.json exists, this is a workflow-enabled project:
   • Resolve the docs repo path: if docsRepo is "." this IS the docs repo; otherwise run pwsh .claude/skills/tool-worktree/scripts/resolve-repo.ps1 <docsRepo> to find the actual path (works in bare+worktree, agentsandbox, and normal clones)
   • Note the resolved docs repo path for later reference
   • Skills that generate documents will output to the docs repo
   • The docs.output paths in workflow.json are relative to the resolved docs repo root

Step 4: Create or Append Daily Memory

8. If context/memory/{YYYY-MM-DD}.md does not exist, create it:

# {YYYY-MM-DD}

## Session 1 — {HH:MM}

### Goal
[Pending — awaiting user goal]

If it already exists (second+ session today), append a new session header:

## Session {N} — {HH:MM}

### Goal
[Pending — awaiting user goal]

Step 5: Freshness Check

9. Check modification dates on all files in context/:
   • If any file is older than 30 days, flag it: "Your [filename] is from [date]. Want to refresh, or keep going?"
   • If context/memory/ has daily files older than 30 days that aren't archived, suggest archiving them to context/memory/archive/.

Step 6: Documentation Status (if workflow-enabled)

10. If workflow.json exists and has a docsRepo path:
    • Check what docs were recently created or modified in the docs repo (last 7 days: ADRs, RFCs, design docs, etc.)
    • Check recent session notes for decisions that might need an ADR
    • If any recent decisions lack a corresponding ADR, note it: "Decision from [date] about [topic] — should this be an ADR?"
    • Briefly list available document types: "You can generate PRDs, RFCs, ADRs, design docs, C4 diagrams, runbooks, post-mortems, and spike reports."

Only surface documentation status if there's something actionable. Don't list document types every session — only on first session or if asked.

Step 7: Skill Inventory

11. List available skills from .claude/skills/ (names and one-line descriptions only). Note any skills that were recently added or modified.

Step 8: Greet

12. Greet the user briefly. If recent daily files contain relevant context (open items, ongoing work), mention it. Keep it to 2-3 sentences max.

Example: "Morning. Yesterday you were working on the event sourcing migration — left off at the projection rebuild. You also made a decision about caching strategy that might need an ADR. Ready to continue, or something new?"