dream-rsi inference

Run this as a background subagent on the smallest available model. The subagent reads files and returns one JSON object. It does not edit files or dispatch verbs.

Inputs:

• observations.json files: <repo>/.gm/dream-rsi/<session>/observations.json, each a JSON array of up to the newest 256 records with dispatch_id, verb, fingerprint, exit_code, gate_drift (bool), prd_open_count, mutable_open_count, quality (0..1), ts, reply_sha256, and optional lean_node. Orchestration verbs and dream-* dispatches are not recorded.
• The gm graph: skills/dream-rsi/gm-graph.json (nodes, edges; node kind is principle, gate, tension or terminal -- there is no phase node; edge kind is forward or backreference). A node is a book when its book is non-empty; book_index names each anchor book and the node that carries it.
• Phase history: .gm/turn-state.json phase_history.
• Walk evidence: .gm/lean-walk/<NODE>.json, one subagent record per node (node, applied, witness).

Part A: verb strategy

1. Read every observation file. Count records. If there are fewer than INSUFFICIENT_RECORD_THRESHOLD across all sessions, return an Insufficient for Part A.
2. Per verb, compute count, failure rate (exit_code != 0), gate-drift rate, and mean quality.
3. Treat quality with caution. Until the scoring fix in rs-plugkit lands, quality mostly reflects the open PRD/mutable backlog rather than the dispatch outcome. Rank by failure rate weighted by count, not by quality.
4. Some non-zero exits are valid answers, not failures. A grep exit 1 with no matches is a no-match result. The agent shall not propose a change for a verb until its failure exits are shown to be real errors.
5. For up to MAX_PART_A_PROPOSALS candidates, write one proposal with the verb, the change, the counts and rates as evidence, and a measurable target on the same log.

Part B: gm-graph traversal