amitabhainarunachala/dharma_swarm/chetana/claude_code_plugin/chetana/skills/chetana

chetana — connective tissue for the grand memory system

You have access to chetana (Sanskrit: चेतना, "consciousness") — a Python package that connects Dhyana's five distributed knowledge substrates into one operational PKM. It is installed as a plugin and accessible via:

• CLI: python -m dharma_swarm.chetana.cli <subcommand>
• Hooks: SessionStart, Stop, SessionEnd, PreCompact, SubagentStop fire automatically every session
• Slash commands: /chetana-status, /chetana-revive, /chetana-gap-scan, /chetana-ingest, /chetana-palace

When to invoke chetana

| User signal | Move | |---|---| | "what's stale?" / "what needs re-verification?" | chetana decay | | "revive my wiki" / "re-integrate stale atoms" | chetana revive --all (proposal); --apply to write | | "what gaps are in the wiki?" / "what should I research?" | chetana gap-scan | | "ingest this PDF / clip / note" | chetana ingest <source> --kind <pdf|webclip|note> | | "show me the memory palace" / "where do my atoms live?" | chetana palace | | "promote this staged atom" / "approve atom X" | chetana promote <staged_path> | | "what did I capture last session?" | check ~/.dharma/sessions/captures/daily/<today>/ | | "show me chetana state" | chetana status |

The grand-memory architecture

Five layers, each chetana CLI surfaces one:

| Layer | What | chetana surface | |---|---|---| | L0 atoms | frontmattered markdown | ingest writes staged; promote writes trusted | | L1 graph | bidirectional links over memory MCP + gitnexus + contextplus + catalytic | query | | L2 PARA | dynamic projections by Projects/Areas/Resources/Archives | palace --para | | L3 palace | 10 Pillars rooms, JSON Canvas | palace | | L4 governance | TelosGatekeeper + KernelGuard + provenance schema | every promote and revive --apply routes through it |

Continuous evolution cycle (already wired)

1. Every SessionStart: hook surfaces stale atoms + top wiki gaps + recent captures into the systemMessage. You start each session knowing what's stale, what's missing, what was captured.
2. Every Stop: hook ingests the just-finished JSONL into staged atoms in ~/.dharma/knowledge/staging/<date>/. Nothing auto-promotes — review before trusting.
3. Every PreCompact: hook drops a checkpoint to ~/.dharma/sessions/captures/in_flight/ so context loss doesn't lose the session.
4. Every SubagentStop: hook records subagent stops to a daily manifest.

Stale → revive, not exile

When an atom passes stale_after, the default move is REVIVE, not quarantine. Revival means: read the atom, scan corpus for new neighbors / new backlinks / questions now answered, propose patches, re-sign axiom signature, append a revival_chain entry to provenance. Quarantine is opt-in last resort (chetana decay --quarantine).

Reference paths

• chetana code: /Users/dhyana/dharma_chetana/dharma_swarm/chetana/ (worktree on feat/chetana-grand-memory)
• chetana README: /Users/dhyana/dharma_chetana/dharma_swarm/chetana/README.md
• Staging area: ~/.dharma/knowledge/staging/<date>/
• Trusted atoms: ~/.dharma/knowledge/wiki/concepts/
• Quarantine: ~/.dharma/knowledge/quarantine/
• Session captures: ~/.dharma/sessions/captures/daily/<date>/
• Hook log: ~/.dharma/sessions/captures/chetana_hook.log

When NOT to invoke chetana

• For ad-hoc Python scripts not tied to memory ops — chetana is not a general utility.
• For touching dharma_swarm runtime (the daemon, the gates) — that's separate.
• For chat history of the current session — chetana captures on Stop, not in-flight.

Quick reference

# read-only checks
chetana status
chetana decay
chetana gap-scan
chetana palace

# write operations (require gate check)
chetana ingest "raw text or path" --kind note --title "Atom title"
chetana promote ~/.dharma/knowledge/staging/2026-04-27/<id>.md
chetana revive --all --apply --reviewer dhyana