feat: hand back the colliding note, not just its id (0.7.3)

[vib795]

The last piece of the capture pipeline, and the one that survived contact with brief.

The problem

A title collision reported the other note's id and stopped:

warning: title collision: session-auth reads the same as existing use-sessions;
         set contradicts or merge them

Deciding what to do with that — merge a duplicate, supersede a changed claim, or link a genuine contradiction — needs the other note's words. Getting them meant a second agent-memory get. On a biller that charges per prompt rather than per tool call, that is another request for something write already had in hand.

Now

created session-auth [decision]
warning: title collision: session-auth reads the same as existing use-sessions [decision] Chose server sessions over JWT
    Why: revocation had to take effect immediately.
    Rejected: short-TTL JWT, because logout would lag by the TTL.
    Implemented in src/auth/session.js:42, with the store in Postgres so a restart
    does not sign everyone out.
    -> update use-sessions by id, or set supersedes or contradicts on session-auth.

Fixed in the turn that found it. --json gains a collisions array with existing, existingType, existingTitle, existingArchived, excerpt, truncated.

One extra row is read, and only when a collision actually occurred — bodies are still not loaded for the whole store to describe the one that collided.

The excerpt is cut on a word or line boundary, for the same reason the digest sheds whole items rather than trimming characters: text that stops mid-word reads as corrupted. When clipped, it says where the rest is:

    ... agent-memory get long-note for the rest

Why this matters more than it looks

This is precisely what content-hash dedup cannot catch. compact merges notes whose content is identical; two notes making one claim in different words are not identical, only synonymous. No amount of later compaction repairs that — the moment to catch it is the write.

Not built, deliberately

write --dry-run was on the Tier 3 list. It would turn one call into two, and brief already answers "what is already here" before composing — the same question, asked earlier and cheaper. Adding it would have made the turn economics worse, not better.

Verification

• 111 tests, up from 109 — two new, covering the returned body, the self-update case that must stay silent, and boundary clipping
• skills/remember/SKILL.md updated: decide in this turn, do not run get for what you were handed
• Zero dependencies; all versions synced to 0.7.3

🤖 Generated with Claude Code

https://claude.ai/code/session_014fSLBRUVVhAauWDuzJM4mc

[coderabbitai]

Warning

Review limit reached

Next included review available in 37 minutes.

View limit details

Limit details: You’ve used the included review currently available.

You've used all free OSS reviews for now. Wait for the free limit to reset to keep reviewing this public repository.

Learn how review limits work.

Review configuration:

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro Plus

Run ID: 485e302f-7719-4315-ae64-0c9074ca316f

📥 Commits

Reviewing files that changed from the base of the PR and between 7593c22 and 9a5229c.

⛔ Files ignored due to path filters (1)

• package-lock.json is excluded by !**/package-lock.json

📒 Files selected for processing (8)

• .claude-plugin/marketplace.json
• .claude-plugin/plugin.json
• README.md
• package.json
• skills/remember/SKILL.md
• src/cli.js
• src/config.js
• test/integration.test.js

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}