janbjorge/rekal/rekal-hygiene

Find and fix problems in the rekal database. Every change requires explicit user approval. Never auto-delete. Never auto-modify.

Step 1: Health overview

rekal health

Report one line:

"142 memories across 3 projects, spanning 6 months."

Step 2: Audit the full store

rekal export

This prints every memory as JSON (id, content, project, tags, timestamps). Read it and group entries that cover the same fact/preference/procedure.

Per finding, classify and propose:

Issue?
├── Duplicates / same topic stored twice
│   └── Identify the most complete version, propose replacing the others:
│       memory_store(content="<best content>", replaces="<older id>")
│
├── Contradiction: one outdated, one current
│   └── Propose: memory_store(content="<current content>", replaces="<outdated id>")
│
├── Contradiction: unclear which is correct
│   └── Ask user: "Which is correct? [A] or [B]?"
│
├── Project-specific content but project missing
│   └── Propose: memory_store(content="<same content>", project="<name>", replaces="<id>")
│
├── Content < 20 characters or meaningless alone
│   └── Propose: reword with more context (store with replaces), or delete if worthless
│
└── No issues → skip

Skip memories created < 24 hours ago: too fresh to judge.

Step 3: Present action plan

Compile ALL proposals into one summary:

Hygiene report:

- 3 duplicate clusters: 8 memories → 3

- 2 contradictions: 1 replace, 1 needs your call

- 2 unscoped memories to re-scope

Approve all? Or review individually?

Wait for explicit approval. Do NOT execute anything until user approves.

Step 4: Execute approved changes

After user approves (all or specific items):

1. Run each approved operation:

• memory_store(..., replaces=<id>) for duplicates, corrections, re-scoping
• memory_delete only for worthless entries (user explicitly approved)
• rekal prune --older-than-days N --yes for approved bulk cleanup

1. Report what was done, one line per action
2. Run rekal health again, report improvement:

"Done. 142 → 134 memories. 3 duplicate clusters resolved."

Safety rules

These are hard rules. No exceptions.

• Never auto-delete. Every deletion requires user approval.
• Never auto-modify. All changes proposed first, executed after approval.
• Skip < 24h old memories. Too fresh to judge quality.
• Replace over delete. replaces keeps the topic covered. Delete only genuinely worthless entries.
• No new memories. This skill cleans. /rekal-save stores.
• No bulk approve without listing. Always show what will change before asking for approval.

Large databases (500+ memories)

Do NOT audit everything in one pass. Prioritize in this order:

1. Contradictions: highest impact on trust
2. Topics appearing >= 3 times in the export: most duplicate accumulation
3. Newest 30 days of entries: freshest quality issues

After completing priority items, ask:

"Covered contradictions and top duplicate clusters. Want me to continue with a deeper audit?"