What To Do

Add ^tr IDs to untagged memory entries in managed files. This is safe to run multiple times - it only tags entries that don't already have IDs.

What To Do

1. Identify Managed Files

Scan these locations for single-line list item entries:

• CLAUDE.local.md (working memory)
• memory/registers/*.md (all register files)
• memory/archive/**/*.md (archived entries, if they exist as list items)

Do NOT scan memory/daily/*.md - daily logs are not managed by the remove gate.

2. Parse Entries

An entry is exactly:

• A line starting with - (dash + space)
• Single line only (the entire entry is on one line)
• NOT inside a fenced code block (``` markers)
• NOT a placeholder like - [None captured yet] or - [None yet] or - [not yet set]

Already tagged: If the line ends with ^tr followed by exactly 10 lowercase hex characters, it already has an ID. Skip it.

Malformed tag: If the line ends with ^ followed by other text that doesn't match the tr[0-9a-f]{10} pattern, warn and skip it. Do not attempt to repair.

3. Show Preview

Before making any changes, show the user what will happen:

recall-init-ids — preview
━━━━━━━━━━━━━━━━━━━━━━━━━

Files scanned: [N]
Entries found: [N] total
  Already tagged: [N]
  Will be tagged: [N]
  Skipped (malformed): [N]

Examples:

  CLAUDE.local.md:
    Before: - Don't blast into writing code unless explicitly told to
    After:  - Don't blast into writing code unless explicitly told to ^tr8f2a1c3d7e

  registers/preferences.md:
    Before: - Prefers tabs over spaces
    After:  - Prefers tabs over spaces ^tra4b2c1d9e0

Apply? [y]es / [c]ancel

Show up to 5 before/after examples, spread across different files.

If no untagged entries are found, report that and exit:

All [N] entries already have IDs. Nothing to do.

4. Generate IDs

For each entry to be tagged:

1. Generate an ID: tr + 10 random lowercase hex characters (use crypto-secure randomness)
2. Check for collisions against:
   • All existing inline IDs found during parsing
   • All IDs already generated in this run
   • All keys in memory/.recall/metadata.json (if it exists)
3. If collision, regenerate. (Extremely unlikely with 10 hex chars, but be safe.)
4. Append ^[id] to the end of the line

5. Apply Changes

Only after user confirms:

1. Rewrite each file with IDs appended to eligible entry lines. Do not modify any other lines in the file.

2. Create or update memory/.recall/metadata.json:

   • Create memory/.recall/ directory if it doesn't exist
   • For each new ID, add an entry:

     {
       "tr8f2a1c3d7e": {
         "created_at": "2026-02-12T10:15:30Z",
         "last_reviewed_at": "2026-02-12T10:15:30Z",
         "pinned": false,
         "snoozed_until": null,
         "status": "active",
         "tier": "working"
       }
     }
     

   • Set tier based on file location:
     • CLAUDE.local.md -> "working"
     • memory/registers/* -> "register"
     • memory/archive/* -> "archive"
   • For entries with existing inline IDs that are missing from metadata.json, add them without changing the file
   • Write JSON with sorted keys for readable diffs

6. Summary

recall-init-ids — complete
━━━━━━━━━━━━━━━━━━━━━━━━━━

Tagged: [N] entries across [M] files
Metadata: [N] entries in memory/.recall/metadata.json

Files modified:
  CLAUDE.local.md          [N] entries tagged
  registers/preferences.md [N] entries tagged
  ...

You can now run /recall-maintain for pressure-based memory cleanup.

Important Notes

• This command is idempotent. Running it again will only tag NEW untagged entries.
• If a user manually removes an ID from an entry, running this command will re-tag it with a NEW ID (losing pin/snooze/review history for that entry).
• Metadata.json is the source of truth for pin/snooze/review state. The inline ID is the link between the entry text and its metadata.