source-command-compress-memory

name	source-command-compress-memory
description	Archive finished tasks, snapshot active memory, keep AGENTS.md under 40k chars

source-command-compress-memory

Use this skill when the user asks to run the migrated source command compress-memory.

Command Template

/compress-memory

Perform a memory compression pass on this project. Work through the four steps below in order. Report results at the end — do not ask for confirmation between steps.

---

Step 1 — Check AGENTS.md size

Read AGENTS.md and count its characters (wc -c AGENTS.md).

• If ≤ 40,000 chars: note the size and continue.
• If > 40,000 chars: the root file has grown beyond the lightweight-index contract. Identify which sections were added directly to AGENTS.md instead of to a .Codex/rules/ file, move them, and replace with a @.Codex/rules/<file>.md pointer. Target: back under 40,000 chars.

---

Step 2 — Archive finished tasks from roadmap-todo.md

Read .Codex/rules/roadmap-todo.md.

Find all items that are unambiguously done — marked with strikethrough (~~...~~), labelled RESOLVED, IMPLEMENTED, DONE, or already present in the codebase with no remaining action.

Move them to .Codex/rules/archive.md under a heading:

## Archived <YYYY-MM-DD>

Append to the file if it already exists; create it if it does not.

Remove the moved items from roadmap-todo.md. Do not remove items that are partially done or still have sub-tasks open.

---

Step 3 — Snapshot active memory

Read MEMORY.md (the project memory index at /Users/jens/.Codex/projects/-Users-jens-develop-geberit-aquaclean/memory/MEMORY.md).

For each linked memory file:

• If the memory duplicates something already captured in .Codex/rules/ verbatim, remove the memory file and its MEMORY.md index entry.
• If the memory refers to a fact now embedded in AGENTS.md or a rules file (e.g. a feedback rule that became a mandatory rule), remove it from MEMORY.md and its file.
• If the memory is still load-bearing (non-obvious, not derivable from the current code or rules), keep it exactly as-is.

After pruning, write a one-line snapshot entry to .Codex/rules/archive.md:

## Memory snapshot <YYYY-MM-DD>
Active memory entries: N  |  Pruned: M  |  Snapshot taken by /compress-memory

---

Step 4 — Report

Print a concise summary:

/compress-memory results
========================
AGENTS.md:         <before> chars → <after> chars  (limit: 40,000)
Tasks archived:    N items moved to .Codex/rules/archive.md
Memory pruned:     M entries removed from MEMORY.md
Files modified:    list them

If nothing needed changing in a step, say so explicitly (e.g. "AGENTS.md: 3001 chars — within limit, no change").