菜单
measure whether a /dream pass helps, hurts, or ties against no consolidation. usage: /learning-loop:dream-eval --mode={single|control|repeated} [--passes=N] [--mine]. single scores pre/post on the live corpus; control forks two clones and dreams one; repeated dreams a clone N times and plots the drift + content-survival curve.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Deep research with the local librarian doing the token-heavy middle. Scope on Claude, Search+Fetch+Extract on the local Ollama model, adversarial Verify + Synthesize back on Claude. Usage: /learning-loop:research "<question>". Falls back to Claude-native WebSearch when the librarian is unavailable or sub-tier.
End-of-session consolidation. Usage: /learning-loop:reflect (no args). Reviews conversation, extracts learnings, routes to auto-memory or vault, cross-links projects, promotes inbox notes. Run after substantial work sessions.
Consolidate auto-memory between sessions. Usage: /learning-loop:dream (no args). Four-phase cycle: Orient, Gather Signal, Consolidate, Prune Index. Seven operators: MERGE, RESOLVE, ABSTRACT, COMPRESS, PRUNE, LINK, DATE NORMALIZE. Rebuilds MEMORY.md.
Cross-store correction for a retracted or updated belief. Usage: /learning-loop:rewrite "old pattern" "new pattern" [reason]. Searches vault, auto-memory, and episodic history; presents an impact map for triage; executes user-approved changes; records a supersession so future episodic searches get annotated.
Diagnose your learning-loop installation. Runs health checks, presents issues, offers per-fix remediation, re-runs each check after the fix to confirm. Safe to run anytime; only makes changes you approve. Pass --redact to scan plugin data files for leaked credentials instead.
Show all learning-loop commands with usage and modifiers. Usage: /learning-loop:help (no args).
| name | dream-eval |
| description | measure whether a /dream pass helps, hurts, or ties against no consolidation. usage: /learning-loop:dream-eval --mode={single|control|repeated} [--passes=N] [--mine]. single scores pre/post on the live corpus; control forks two clones and dreams one; repeated dreams a clone N times and plots the drift + content-survival curve. |
wraps /dream without changing it, and answers three questions with a number: did this pass help, is /dream better than no consolidation at all, and does it degrade the corpus over repeated passes. report-only. control mode is the one that matters most: on a low-redundancy corpus the published prior says consolidation only ties raw retrieval, so proving it locally is the point.
| Input | Mode | Passes | Mine first |
|---|---|---|---|
/learning-loop:dream-eval | single | n/a | no |
/learning-loop:dream-eval --mode=control | control | n/a | no |
/learning-loop:dream-eval --mode=repeated --passes=5 | repeated | 5 | no |
/learning-loop:dream-eval --mine | (mode) | yes |
node "${CLAUDE_PLUGIN_ROOT}/scripts/marker.mjs" lock-acquire dream. exit 0 means proceed. exit 1 means another /dream or dream-eval run is active (or a crashed one less than an hour old), so stop and tell the user. exit 2 means a usage or install error, report the stderr message and abort.probes.jsonl is absent and --mine was not passed, stop and say: "no probe corpus found. run with --mine first." (release the lock before stopping, see step 6).--mine, run the probe miner (forward + reverse) and persist to probes.jsonl.node "${CLAUDE_PLUGIN_ROOT}/scripts/marker.mjs" lock-release dream.