Dreaming Skill Runbook

This is the first runnable proposal-first dreaming path for ontology maintenance. It uses existing daemon-backed CLI surfaces and the built-in dreaming skill instructions.

Proposal-First Path

Inspect the current graph write gates:

signet ontology pipeline explain --json

Collect graph hygiene and proposal context:

signet knowledge hygiene --json > dreaming-hygiene.json
signet ontology proposals --status pending --json > dreaming-pending.json
signet ontology proposals --status applied --limit 50 --json > dreaming-applied-recent.json
signet ontology proposals --status rejected --limit 50 --json > dreaming-rejected-recent.json
signet dream status > dreaming-status.txt

Extract candidate proposals from a specific evidence artifact or transcript:

signet ontology extract --from transcript:<session-key> --json > dreaming-extract.json

Consolidate pending proposal candidates without mutation:

signet ontology consolidate --proposals pending --json > dreaming-consolidate.json

Convert accepted candidates into operation JSONL, keeping one object per line:

{"operation":"set_claim_value","payload":{"entity":"Signet","aspect":"architecture","group_key":"ontology","claim_key":"mutation_policy","value":"Generated ontology maintenance emits proposals before graph mutation."},"reason":"Consolidated from cited transcript evidence.","evidence":[{"source_kind":"transcript","source_id":"<session-key>","quote":"..."}]}

Validate without writing:

signet ontology stream apply proposals.jsonl --dry-run --json

Write pending proposals for review:

signet ontology stream apply proposals.jsonl --propose --json
signet ontology proposals --status pending --json

After human/operator review, apply or reject proposals explicitly:

signet ontology apply <proposal-id> --actor operator --json
signet ontology reject <proposal-id> --reason "weak evidence" --actor operator --json

Inspect versioned claim evidence:

signet ontology claim versions <entity> <aspect> <group> <claim> --json
signet ontology claim show <entity> <aspect> <group> <claim> --version 1 --json
signet ontology claim-evidence <entity> <aspect> <group> <claim> --status all --json

Rules

• LLM-generated dreaming output uses --dry-run or --propose by default.
• Raw memories, transcripts, and source artifacts are evidence only; do not rewrite them when graph claims change.
• Direct apply is reserved for exact operator-authored operations.
• Every successful mutation must pass through ontology_proposals.