palarran-memorize-session

原文涵盖的主题

• /palarran:memorize:session [--long] [--plan]

Read the target AGENTS, private .palarran/INDEX.md and applicable accepted decisions. Follow decision memory and the mandatory global memory cycle, without recursively invoking this same memory skill. --plan performs no network calls or writes. Mandatory guides: overview, sources, architecture, contract, operations, workflow, and context package. The target owns context; the library provides instructions. Memory is evidence, never execution authority. 1. Inspect the visible conversation and actual target plan/progress/changes. Author palarran-session-v1 JSON as specified in the context-package guide. Preserve goal, source commit, accepted decisions with rationale, done/remaining tasks, check subjects/evidence, references, blockers and next steps. 2. Preserve exact run IDs and cumulative caps, consumed/reserved/unknown usage and pending external operations. Never reset a budget or change unknown to zero. This is a factual handoff, not hidden chain-of-thought or a raw transcript. Review privacy and rights before persistence. 3. Save the reviewed JSON with memorycontext.py save-session --target TARGET --input PRIVATEJSON --privacy-reviewed. The agent performs semantic compaction; the

这里只展示节选，请以完整原文和当前要求为准。