memory-compact

The full skill.

Original instructions from the publisher’s SKILL.md

# Memory Compact

You reduce memory bloat without losing the reasoning future agents need.

## Workflow

1. Identify scope: project file, global file, or whole store.
2. Check line counts and active budgets.
3. Classify entries as keep, merge, summarize, archive, delete-candidate, or unsafe.
4. Preserve active decisions, rationale, alternatives, revisit triggers, and provenance.
5. Merge duplicate or overlapping entries.
6. Move old session detail to `archived/` after producing a rollup summary.
7. For global memory, prefer deletion or archival over long summaries.
8. Update routing/index files after compaction.
9. Log project file outputs in `docs/skill-outputs/SKILL-OUTPUTS.md`.

## Global Budgets

- `user-preferences.md` <= 100 lines.
- `global-agent-rules.md` <= 150 lines.
- `reusable-learnings.md` <= 200 lines.
- `global-index.md` <= 250 lines.
- Active global total target: 500-700 lines.

## Compaction Rules

| Entry type | Action |
|---|---|
| active decision | keep or tighten |
| superseded decision | archive with replacement link |
| repeated session details | roll up |
| low-confidence maybe | keep local or archive |
| obvious general advice | delete |
| sensitive content | invoke `memory-forget` |

## Hard Rules

- Never append to an over-budget file without compacting first.
- Never delete active decisions, revisit triggers, or provenance.
- Never compact global memory without user approval.
- **Shrink guard:** Refuse compaction that removes >50% of indexed entries unless the user explicitly approves — archive superseded content instead of deleting.
- Never increase global memory during compaction.
- If meaning would change, stop and ask the user.

## Output Format

```markdown
Compaction complete
Scope: <project/global/file>
Before: <line count>
After: <line count>
Archived: <count/path>
Deleted: <count/reason>
Preserved decisions: <count>
```

## Example

If `reusable-learnings.md` is 240 lines, merge duplicate lessons, archive stale items, and stop only when it is <= 200 lines.

## Common Rationalizations

| Excuse | Reality |
|--------|---------|
| Skip memory — just code | Next agent loses decisions, blockers, and approved scope. |
| Load every memory file | Read indexes and handoff tail only — bounded context. |
| Global memory for everything | Project memory default; global only when stable and cross-project. |
| External paste → memory | Run secure-* first; transform to agent-authored notes. |

## Verification

- [ ] Correct sub-skill routed with reason
- [ ] No secrets or raw transcripts persisted
- [ ] Files changed listed in Impact Report
- [ ] Security gate noted when external content involved

## Red Flags

- Append to over-budget file without compacting first
- Active decisions or revisit triggers deleted during compact
- Global memory compacted without user approval
- Provenance stripped when merging duplicate entries

## Prune Log
Last pruned: 2026-07-04
- No changes — citation audit passed; content current (improve-skills full pass 2026-07-04)


## Impact Report

After completing, report:
```markdown
Memory compacted: <path>
Line reduction: <before> -> <after>
Budget compliant: yes/no
Entries archived: <count>
Entries deleted: <count>
```