mnemos-capturelisted

# Mnemos Capture Manual counterpart to the `SessionEnd` capture hook. The hook reads the raw transcript and runs it through extraction after the session is over. This skill runs *now*, while you still hold the session in context — which means you can distill what actually mattered instead of leaving a model to mine it out of a 20KiB transcript later. Use it before `/clear`, before a compaction, or any time the user says a conclusion is worth keeping. ## Choosing a path Prefer the **MCP tools** (`mcp__mnemos__*`) — they work identically against a local and a hosted brain. Fall back to the **CLI** only when no `mcp__mnemos__*` tool is available or every call errors; the CLI resolves its own DSN and may target a different brain than the MCP server is serving. ## Step 1 — Decide what is worth keeping This is the whole job. The brain is only as useful as it is uncluttered, and every junk claim dilutes recall for every future session. **Capture:** - Decisions, with the reasoning and the alternatives rejected. - Facts discovered about the system that were not obvious from the code — behaviors, constraints, version quirks, why something is the way it is. - Corrections: things believed at the start of the session that turned out to be wrong. These are the highest-value claims in the brain, because they overwrite something that is already in there misleading people. - Preferences and constraints the user stated. **Do not capture:** - Your own narration ("I'll now read t