palarran-memory-learn

原文涵盖的主题

• /palarran:memory:learn

Shared cycle: before work and after checks follow global memory: bounded recall → applicability → sanitized verified learn/dedup. This memory scope is explicitly authorized; other profile restrictions remain. --plan permits no network/publication; outages do not block the main task. Mandatory guides before decisions: README.md sources.md architecture.md contract.md operations.md workflow.md Decision memory 1. Read target INDEX/accepted ADRs/plan revision/final evidence. Select at most 3 lessons including failures/anti-patterns; test passes do not establish universal patterns. 2. Create immutable cases using packages/memory/knowledge/templates/case.example.json: problem/rootcause/solution/applieswhen/limitations/source/evidence. Separate observations/hypotheses; never set privacyreviewed without checking contents/rights. No raw session/log/env/customer-data imports. 3. Find existing cases through bounded search/get; check duplicates/conflicts/originals. Identical content deduplicates by hash; similar cases never overwrite history. 4. --plan writes drafts/rationale only, no service upload. Otherwise authorized memory-write propose creates candidates; agent credentials grant no promot

这里只展示节选，请以完整原文和当前要求为准。