ownmem

name	ownmem
description	Recall this repository's OwnMem local memory before changing code, and keep it healthy. Use when a repository contains .ownmem/, when past debugging lessons could apply ("have we hit this before", "why is it done this way"), or when the user mentions ownmem, project memory, or recalling across sessions.

OwnMem recall and daily use

OwnMem is this repository's local, deterministic memory. Recall runs offline — no model calls, no network, no query-time token cost — and every memory is plain Markdown inside the repository.

If the repository has no .ownmem/ directory, OwnMem is not installed here; use the init skill from this plugin first.

Recall before changing code

Run one recall for the problem, path, or symbol you are about to touch:

npx ownmem recall -- "problem, path, or symbol"

For natural-language symptoms, pass two or three phrasings in one call and let the deterministic fusion rank them:

npx ownmem recall --multi -- "symptom in plain words" "root-cause term" "english symbol"

Open a hit's full topic file before relying on it, then verify against live code: memories record what was true when they were written.

The three feedback streams

Three different questions, three separate local files, three separate denominators. Nothing here is uploaded, and no stream may stand in for another.

1. Retrieval verdict — did recall return the right thing?

npx ownmem recall --feedback correct -- "query that matched"
npx ownmem recall --feedback retrieval_miss --expected memory_name -- "query that should have matched"

retrieval_miss means the right memory is active but fell outside the top-k, so it must name it. coverage_gap means no right memory exists yet, so it never takes --expected. wrong, stale and conflict may name one.

2. Outcome receipt — what happened after a memory was used? Record it only when the user or the host actually confirmed it; an agent's own opinion is refused here on purpose. Only the digest of the confirming statement is stored, never its text.

npx ownmem outcome --memory memory_name --outcome applied \
  --confirmed-by user --confirmation "yes, that fixed it"

This is the only honest measure of actual application. Until receipts accumulate, a full-text open still means only that a body was read.

3. Weak self-attribution — at the end of a turn in which memory was injected, and only when a memory clearly helped or clearly misled you:

npx ownmem attribute --memory memory_name --label useful

Record nothing when the turn was neutral. Because the sample selects itself, these labels are counts and never a rate: an unlabelled turn is unknown, not neutral.

Keep the memory healthy

npx ownmem audit              # schema, quota, boundary, and duplicate gates
npx ownmem report --since 7d  # adoption and recall-quality report
npx ownmem dashboard --open   # local Web console

Write a new memory only for lessons that reading the code cannot recover — upstream quirks, toolchain traps, timing races. Keep entries small and let npx ownmem audit enforce the growth quota.