# weekly-memory-cleanup: a Claude Code skill by Scalably

Canonical: https://scalably.io/skills/weekly-memory-cleanup
Source: https://github.com/scalably-io/agent-skills/blob/main/skills/weekly-memory-cleanup/SKILL.md (MIT, v1.0.0)
Plugin: agent-ops · Provenance: derived from a Scalably production skill, source commit 7242274a, verified 2026-09-07
Integrity: sha256 of the raw SKILL.md served at https://scalably.io/skills/weekly-memory-cleanup.md is 318d1193aabe8243e027a6ebfc9da76301e160f95902551a2a778e46c4c09824. Verify: curl -s https://scalably.io/skills/weekly-memory-cleanup.md | shasum -a 256
This is the machine-readable representation of the page at the canonical URL. Same facts, denser format. The raw SKILL.md is the installable artifact.

## Direct answer
Runs the memory-system skill's Sunday deep pass: the nightly routine first, then archiving the finished week, deduping and topic-merging learned-corrections, reviewing the tree for staleness, cross-file deduping, and rebuilding the index from scratch. Use for weekly memory cleanup, a Sunday memory cleanup, a weekly deep pass, archiving the week, or a memory audit. Not for the nightly pass alone: that's the dream skill; this skill runs it first and then goes further.

## Install
- Claude Code: `/plugin marketplace add scalably-io/agent-skills` then `/plugin install agent-ops@scalably-agent-skills`, then run `/agent-ops:weekly-memory-cleanup`.
- Codex or any Agent Skills runtime: `git clone https://github.com/scalably-io/agent-skills && cd agent-skills`, then `cp -R skills/weekly-memory-cleanup .agents/skills/`. Skills that link to sibling skills expect those siblings to be alongside them: copy the whole `skills/` directory instead of a single folder to keep those links working.
- Raw file: `curl -s https://scalably.io/skills/weekly-memory-cleanup.md -o SKILL.md`

## Requirements
- The `memory-system` skill (same plugin): defines the tree layout and the exact Sunday routine this skill runs, including `scripts/check-memory.sh` and `scripts/archive-week.sh`.
- A week of `./memory/daily/*.md` logs (from the `daily-log` skill or equivalent) and an existing `./memory/` tree, ideally maintained nightly by the `dream` skill.

## Inputs and outputs
| | |
|---|---|
| Input | The full `./memory/` and `./rules/` tree, plus this week's `./memory/daily/*.md` |
| Output | `./memory/weekly-archive/<year>-W<nn>.md` (the finished week), a reset `./memory/weekly-summary.md`, a deduped `./rules/learned-corrections.md` + `./memory/reference/corrections-log.md`, a rebuilt `./memory/index.md`; a 3-5 line report appended to `./memory/consolidation.log` |

## Reliability
- Release v1.0.0. Structural test suite and `claude plugin validate` run on every push: https://github.com/scalably-io/agent-skills/actions/workflows/validate.yml
- Private client-data scan before every release. Last end-to-end verification: 2026-09-08.
- Derived from production, genericized by hand, reviewed per skill and as a whole before publish.

## Full skill
## What it does

Runs the `memory-system` skill's Nightly routine first, then its deeper Sunday pass: archives the finished week into `weekly-archive/`, dedupes and topic-merges `rules/learned-corrections.md`, reviews the tree for staleness, cross-file dedupes, and rebuilds `index.md` from scratch. This is the enforcement layer: whatever drifted during the week (duplicate rules, contradictions, stale "Now" sections, cap overruns) gets fixed here, not deferred to the following week.

## Requirements

- The `memory-system` skill (same plugin): defines the tree layout and the exact Sunday routine this skill runs, including `scripts/check-memory.sh` and `scripts/archive-week.sh`.
- A week of `./memory/daily/*.md` logs (from the `daily-log` skill or equivalent) and an existing `./memory/` tree, ideally maintained nightly by the `dream` skill.

## Inputs and outputs

| | |
|---|---|
| Input | The full `./memory/` and `./rules/` tree, plus this week's `./memory/daily/*.md` |
| Output | `./memory/weekly-archive/<year>-W<nn>.md` (the finished week), a reset `./memory/weekly-summary.md`, a deduped `./rules/learned-corrections.md` + `./memory/reference/corrections-log.md`, a rebuilt `./memory/index.md`; a 3-5 line report appended to `./memory/consolidation.log` |

## Worked example

Paste into Claude Code with this skill (and `memory-system`) installed, on a Sunday, in a project with a week of dailies and a tree the `dream` skill has been maintaining nightly:

```text
/agent-ops:weekly-memory-cleanup
```

Expected: no chat-facing output beyond a brief confirmation. `./memory/weekly-summary.md` is archived to `./memory/weekly-archive/2026-W36.md` and reset with a fresh Monday header (open items carried forward), `./rules/learned-corrections.md` has duplicate and superseded rules removed or merged, `./memory/index.md` is rebuilt from the tree's frontmatter, and `./memory/consolidation.log` gets a 3-5 line report such as: `2026-09-06 weekly-cleanup: archived W36, merged 2 duplicate rules, retired 1 superseded rule, compressed 3 matured rules, check OK, profile 41/80, index 22/50, weekly 18/80, LC 34/60`.

## How to schedule this

Claude Code scheduled task (interactive): run `/schedule` and describe the cadence, e.g. "every Sunday at 09:00 run /agent-ops:weekly-memory-cleanup". The task runs the skill in a fresh session.

Cron, from any machine with Claude Code installed:

```bash
0 9 * * 0 cd /path/to/project && claude -p "/agent-ops:weekly-memory-cleanup" >> logs/weekly-memory-cleanup.log 2>&1
```

## Procedure

### 1. Load the contract

Read the `memory-system` skill's Sunday routine section in full before doing anything else. It is the single source of truth for the routine, and it explicitly includes running the Nightly routine first.

### 2. Run the nightly routine, then the Sunday deep passes

Follow the Sunday routine phase by phase, exactly as defined in `memory-system`:

1. Run the Nightly routine (orient, read recent dailies, file facts, integrate corrections, update the week, refresh the index, verify).
2. **Archive the week**: run `scripts/archive-week.sh`; carry forward still-open items into the new week's "Open items".
3. **Learned-corrections deep pass**: delete verbatim duplicate rules; topic-merge rules on the same subject into one line (move ancestors to `## Retired` in the corrections log with a superseded-by note); compress matured rules (promoted ≥14 days ago, no re-violation this week) to one-liners, moving examples to the log; resolve any contradiction to the latest rule and log the resolution.
4. **The 80% rule**: any injected file above 80% of its cap gets consolidated this pass, before it can break the cap mid-week.
5. **Tree hygiene**: for every tree file: `updated` older than 60 days → verify its `## Now` against recent dailies, restamp or move stale state out; files over 150 lines → split; orphan files → index them or fold them into an existing file.
6. **Cross-file dedupe**: a fact found in two homes: keep the one the rubric assigns, replace the other with a pointer.
7. **Rebuild `index.md`** from the tree's frontmatter (every file, correct summaries, nothing extra).
8. **Verify + report**: run `scripts/check-memory.sh` until clean; append a 3-5 line report to `memory/consolidation.log`.

### 3. Enforcement stance

You are the enforcement layer for whatever drifted during the week: duplicate rules, contradictions, stale "Now" sections, cap overruns. Fix it this pass; do not defer a violation to next week.

### Hard rules

- Pass `offset`/`limit` on EVERY file Read. Never read session `.jsonl` files.
- Moves, not deletions: archived weeks go to `weekly-archive/`, retired rules go to the corrections log's `## Retired` section. The only thing you may delete outright is a verbatim duplicate line.
- End with the check script clean and a 3-5 line report appended to `memory/consolidation.log` (merged/archived/compressed counts, final sizes of the four injected files). Produce no other user-facing output.