checkpoint（checkpoint 搜索）

Checkpoint — Save Developer Memory

When to Checkpoint

When in doubt, checkpoint — a few extra checkpoints are better than lost context.

• Completed a deliverable — feature slice, bug fix, refactor step
• Made a key decision — architecture, tradeoffs, approach choices that future sessions must follow
• Committed or pushed work — natural checkpoint moment after code lands
• Reaching a stopping point — end of a work session or switching tasks
• Before context compaction — preserve active state (PreCompact hook handles this automatically)
• Found something non-obvious — blockers, root causes, discoveries worth preserving
• User shared requirements/constraints — preserve what future work must honor

Space out checkpoints so each one captures a distinct piece of progress — one per logical milestone is the right cadence.

How to Write Good Descriptions

Your description becomes the markdown body of a .md file. Format it with structure — headers, bullet points, bold, code spans.

The WHAT/WHY/HOW/IMPACT Formula

mcp__goldfish__checkpoint({
  description: "## Implemented JWT refresh token rotation\n\nThe existing single-token approach was vulnerable to token theft.\n\n- **Approach:** Rotate tokens on each use, limiting the attack window\n- **Added:** `RefreshTokenStore` with atomic file writes and 7-day expiry\n- **Tests:** All 12 auth tests passing\n- **Impact:** Unblocks the session management PR",
  tags: ["feature", "auth", "security", "jwt", "refresh-token", "token-rotation", "session-management"]
})

Good vs Bad

GOOD (structured markdown):

## Fixed race condition in checkpoint writes

Concurrent saves could corrupt the daily markdown file.

- **Root cause:** Non-atomic write pattern
- **Fix:** Switched to write-tmp-then-rename with file locking
- **Verified:** Reproduced with parallel test, confirmed fix

BAD (no structure): "Fixed race condition in checkpoint file writes where concurrent saves could corrupt the daily markdown file. Root cause was non-atomic write pattern."

BAD (no context): "Fixed file writing bug"

Structured Fields

Use type to classify your checkpoint for better searchability:

• type: "decision" → include decision + alternatives
• type: "incident" → include context + evidence
• type: "learning" → include impact

All types benefit from symbols, next, and impact.

Tags — Think About Future Search

Tags power fuzzy search recall. Write them for discoverability — how would future-you search for this?

Category tags (1-2):

• Type: feature, bug-fix, refactor, docs, test
• Area: auth, api, ui, database, build
• Status: wip, blocked, discovery, decision

Concept tags (2-5) — the important part:

• Include synonyms and related terms for the core topic
• Include the problem domain, not just the solution
• Think: "what words might I use when searching for this later?"

Example: a checkpoint about implementing retry logic for payment webhooks:

• BAD tags: ["bug-fix", "payments"]
• GOOD tags: ["bug-fix", "payments", "retry", "resilience", "webhooks", "fault-tolerance", "idempotency"]

Example: a decision to use WebSockets over SSE for real-time updates:

• BAD tags: ["decision", "api"]
• GOOD tags: ["decision", "real-time", "websockets", "sse", "push", "streaming", "notifications"]

What Gets Captured Automatically

You don't need to include these — Goldfish captures them:

• Git branch — current branch name
• Git commit — current HEAD short hash
• Changed files — files modified since last commit (.memories/ excluded)
• Timestamp — UTC, always

Focus your description on the MEANING, not the mechanics.