| name | mempal-memory |
| description | Search and record durable engineering knowledge with the mempal CLI, before and after coding work. |
Mempal memory protocol for this eval
This environment has a persistent memory database that survives across tasks, reachable through
the mempal CLI (already installed, on PATH). Run it with your normal shell tool — there is no
MCP server, and you need exactly two commands.
1. Search before you start
mempal search "<key terms of the task>" --json --top-k 5
Run it once, at the start, with the task's key terms (framework, API, error message, symptom).
Empty results are a normal outcome — continue with the task and do not investigate the memory
database. Treat any hit as evidence, not authority: verify it against the current repository
before acting on it.
2. Record before you finish
If the work produced a durable engineering decision or a reusable debugging lesson, write a short
markdown note and ingest it:
cat > "$HOME/.mempal-notes/<short-slug>.md" <<'EOF'
<what was decided or learned, and why — 2-5 sentences, enough for a future task to act on it>
EOF
mempal ingest "$HOME/.mempal-notes" --wing memory-evals
Only mempal search and mempal ingest are needed. Do not call other mempal subcommands
(status, projects, brief, phase3, knowledge*, cowork-*): they print format
specifications and protocol dumps that cost context and tell you nothing about this task.
What not to store
Never store benchmark answers, accepted proposal numbers, hidden-test guesses, raw transcripts,
or task-specific output that would reveal the answer if the same task is run again. Record the
reusable why, not the answer. If nothing reusable was decided, do not invent a note.