AI agent skill · by @mnemon-dev

Mnemon

Persistent memory CLI for LLM agents. Store facts, recall past knowledge, link related memories, manage lifecycle.

Archived 2026-10-23 · Source updated 2026-10-11 · License: Apache-2.0. Check the original repository for the latest version.

SKILL.md instructions

# mnemon

## Workflow

1. **Remember**: `mnemon remember "<fact>" --cat <cat> --imp <1-5> --entities "e1,e2" --source agent`
   - Only exact content repeats are skipped; distinct content is stored and diff suggestions are advisory.
   - To retire a superseded memory, store and verify the new fact, then explicitly run `mnemon forget <old-id>`.
   - Output includes `action` (added/skipped), `semantic_candidates`, `causal_candidates`.
2. **Link** (evaluate candidates from step 1 — use judgment, not mechanical rules):
   - Review `causal_candidates`: does a genuine cause-effect relationship exist? `causal_signal` is regex-based and prone to false positives — only link if the memories are truly causally related.
   - Review `semantic_candidates`: are these memories meaningfully related? High `similarity` alone is not sufficient — skip candidates that share keywords but discuss unrelated topics.
   - Syntax: `mnemon link <id> <candidate> --type <causal|semantic> --weight <0-1> [--meta '<json>']`
3. **Recall**: `mnemon recall "<query>" --limit 10`

## Recall Intent

Keep focused queries and memories in their original language. When the user's
meaning is clear, choose `--intent WHY` (reasons), `--intent WHEN` (timing),
`--intent ENTITY` (what/who), or `--intent GENERAL` (neutral retrieval).
For example: `mnemon recall "<query>" --intent WHY`. The override works in any
language; `--verbose` reports
`meta.intent` and `meta.intent_source` (`auto` or `override`).

Automatic cues cover some forms in English, Mandarin Chinese (simplified and
traditional), Hindi (Devanagari), Spanish, Modern Standard Arabic, French,
Bengali (Bengali script), Portuguese, Indonesian (Latin script), Russian
(Cyrillic), and German. Unrecognized forms use GENERAL; conflicting cues involving
additional languages also use GENERAL. English/Chinese-only scoring is preserved.
This is a lexical heuristic, not full language understanding. See
[the supported forms, script variants, and limits](https://github.com/mnemon-dev/mnemon/blob/master/docs/USAGE.md#recall-intent-detection).

## Commands

```bash
mnemon remember "<fact>" --cat <cat> --imp <1-5> --entities "e1,e2" --source agent
mnemon link <id1> <id2> --type <type> --weight <0-1> [--meta '<json>']
mnemon recall "<query>" --limit 10
mnemon search "<query>" --limit 10
mnemon import --dry-run <file>
mnemon import <file>
mnemon forget <id>
mnemon related <id> --edge causal
mnemon gc --threshold 0.4
mnemon gc --keep <id>
mnemon status
mnemon log
mnemon store list
mnemon store create <name>
mnemon store set <name>
mnemon store remove <name>
```

## Import Historical Chats

When the user asks to import old chats, notes, or exported context, create a
`memory_draft.json` with `schema_version: "1"`, `insights` entries containing
`content`, `category`, `importance`, `tags`, `entities`, and optional
`created_at`, plus optional `edges` using `source_index`, `target_index`,
`edge_type`, `weight`, and `reason`. Run `mnemon import --dry-run <file>`,
then run `mnemon import <file>` only after validation passes. After import,
verify with `mnemon status` and a focused `mnemon search` or `mnemon recall`.
Check the output `errors` field because imports can partially succeed.

## Guardrails

- Main conversation: never run `remember` or `link` directly; delegate to a Task sub-agent.
- Memory sub-agent: you are the leaf writer. Run `mnemon remember` and any justified `mnemon link` commands directly; do not spawn another sub-agent.
- Do not store secrets, passwords, or tokens.
- Categories: `preference` · `decision` · `insight` · `fact` · `context`
- Edge types: `temporal` · `semantic` · `causal` · `entity`
- Max 8,000 chars per insight.