archive-profile-memory

name: archive-profile-memorydescription: Archive or clean up facts from active MEMORY.md into memory-archive/ without deleting them. Use when the user wants to forget, remove, tidy old memory, or free space for new memories.

Use this skill when the user wants to remove facts from active memory without deleting them permanently.

• Use the update-profile-memory skill to add facts to active MEMORY.md.
• Use profile skills for repeatable procedures — not for archiving memory.
• Archived facts live under memory-archive/ and are not loaded into chat automatically. Use search_files or read_file to retrieve them later.

MEMORY.md shape

Active memory is MEMORY.md in the profile workspace:

• A preamble (# Memory Log and ---) must stay intact.
• Facts are bullets under dated sections: ## YYYY-MM-DD followed by - bullet text.

Copy bullet text verbatim from read_file output when editing. Do not paraphrase.

Archive file shape

Archive files live at memory-archive/YYYY-MM.md (year and month of the archive action, UTC).

New archive file template:

# Archived Memory

---

Each archive append starts with metadata comments, then the archived sections:

<!-- archived: 2026-07-12T10:30:00.000Z -->
<!-- reason: user asked to forget old job details -->

## 2026-06-29

- Old preference.

Preserve each bullet's original ## YYYY-MM-DD section header from MEMORY.md.

Workflow

1. read_file MEMORY.md and identify the exact - bullet lines to archive.
2. Confirm targets with the user when the request is ambiguous.
3. Choose memory-archive/{YYYY-MM}.md for the current month.
4. Check whether the archive file exists before calling read_file. With bash, run if test -f memory-archive/{YYYY-MM}.md; then printf 'exists\n'; else printf 'missing\n'; fi in the active profile workspace, replacing {YYYY-MM} with the chosen month. A missing file (or missing memory-archive/ directory) is normal for the first archive:
   • If the result is exists, read_file the archive to preserve its contents.
   • If the result is missing, skip the read and create the archive in step 5.
   • If bash is unavailable, use read_file; treat only a file-not-found result as a new archive. Stop on permission or other errors rather than overwriting an unread archive.
5. Append to the archive first (reduces data-loss risk):
   • If the file does not exist, write_file with the template plus the append block.
   • If it exists, edit_file to append after the existing content (or write_file the full merged content).
6. edit_file MEMORY.md to remove each archived - bullet line.
7. Remove any ## YYYY-MM-DD section that has no bullets left.
8. Re-read MEMORY.md and the archive file to verify the result.
9. If archiving was to free space for new memories, confirm MEMORY.md is under 4096 bytes before retrying with update-profile-memory.

When edit_file reports a missing, duplicate, or overlapping match, read_file again and issue a corrected edit with exact text.

Legacy layout

If you find archives under data/memory-archive/ instead of memory-archive/, use the top-level memory-archive/ path for new writes.