mnemos

Build per-cwd project memory briefing by combining 3 layers (Identity + cwd Sessions + 1-hop wikilink expansion), then synthesizing an evolution-aware narrative (400-600 words). Invoked by SessionStart hook (non-interactive via claude --print) with the current cwd as argument. Output goes to stdout; wrapper writes it to <vault>/.mnemos-briefings/<cwd-slug>.md.

Update <vault>/_identity/L0-identity.md by reading existing identity + new Sessions since last refresh. Synthesizes delta into updated profile, marks revised preferences explicitly, preserves identity-relevant decisions. Invoked from SessionEnd worker subprocess (non-interactive via claude --print) — no arguments needed beyond vault discovery.

Cross-context recall from the Mnemos memory palace. User invokes /mnemos-recall <query> to ask a free-text question; skill searches the palace via MCP, reads top drawers, and synthesizes a narrative answer with wikilink citations. In-session only (no subprocess). Use when cwd-based auto-briefing doesn't cover the query (e.g. working in cwd A but asking about work done in cwd B).

Refines Claude Code JSONL transcripts (~/.claude/projects/) into high-signal Sessions/<YYYY-MM-DD>-<slug>.md notes in the Mnemos vault. Triggers: "/mnemos-refine-transcripts", "refine transcripts", "summarize JSONLs and write to Sessions", "process raw logs", "import Claude Code history to vault", "process unprocessed transcripts". Does NOT call the mnemos API, does NOT call any LLM API — work is done by Claude in this session.