docs: cross-session memory design (P1-4) — kernel-first ownership split + research archive - #45
Draft
Tyan66666 wants to merge 3 commits into
Draft
docs: cross-session memory design (P1-4) — kernel-first ownership split + research archive#45Tyan66666 wants to merge 3 commits into
Tyan66666 wants to merge 3 commits into
Conversation
…it + research archive - docs/cross-session-memory-design.md: full design draft — write (memory_commit at nudge time → one-shot LLM extraction) / store (human-readable markdown memory bank with Ebbinghaus expiry) / readback (resident index + memory_recall, two-timing, delayed readback per Claude Code practice); §6A kernel ownership split (algorithms upstream to acp-kernel, host keeps DSH integration), Phase K → Phase H implementation order - docs/memory-research/: 4 archived research docs — 17 papers (autonomous-memory-papers), 8 products (autonomous-memory-implementations), readback timing 25 sources (memory-readback-timing), Claude Code source verification (claude-code-source-verification) - docs/README.md: index table row for the new design doc - 3 pending items marked at doc top (negative-constraints wording / tool return copy / index schema)
Tyan66666
marked this pull request as draft
August 20, 2026 15:21
…nudge-gated fork), three-tool permission split (memory_write fork-only)
- §3 rewritten: path A (tier-1 block → memory candidate, zero overhead) + path B
(memory_commit → fork extraction); three-tool table — memory_commit (main dialog,
intent only), memory_recall (main dialog, read), memory_write (fork session ONLY,
main dialog has no write permission)
- fork instruction uses {note} as the focus prompt
- §6A kernel/host split updated: three-tool interface semantics → kernel, tool
registration + fork implementation + memory_write persistence → host
- 05-upstream-proposal.md: same dual-path + three-tool structure for the kernel author
…model-decided (nudge + fork only) Path A (engine auto-archiving compressed blocks into the memory bank) was added without user approval and violates decision 3 (no automatic policy). Removed from both the design doc §3 and the upstream proposal; the core insight stays (tier-1 blocks are natural long-term memory candidates) but as background framing only — sedimentation is always a model decision, never an engine side-effect.
Contributor
|
我来协助看一下这个问题,请分析并回复处理结果。 |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
跨会话长期记忆设计文档（P1-4）
模型驱动长期记忆的完整设计稿 + 三轮调研原始资料归档。
设计要点
memory_commit(type, note)表态（纯意图表达）→ 引擎一次性 LLM 调用提取 → 写人类可读 markdown 记忆库（带 citation）→ 主对话继续压缩memory_recall取全文）；Claude Code 生产实践背书（源码核实）score = importance + min(days/30,5) + uniform(0,.5)——打破富者愈富，与保质期同一套字段/acp memory list；导入导出 = 自然语言，模型自主Kernel 归属划分（核心决策）
算法进 kernel、集成留宿主，先改 kernel 再改宿主（§6A）：
调研归档（docs/memory-research/）
待定项（文档顶部标注，实现前补齐）
memory_commitnegative constraints 措辞纯文档 PR，无代码变更，不影响现有 152 tests。