원클릭으로
session-distill
Cursor Composer 会话 Deep Distill:packet → answer-me 验证 → session note → KB。 处理范式与 Grok 一致;禁止无验证的批量晋升。
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
메뉴
Cursor Composer 会话 Deep Distill:packet → answer-me 验证 → session note → KB。 处理范式与 Grok 一致;禁止无验证的批量晋升。
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Distill Codex session JSONL files into review packets, local session notes, and reusable memory candidates. Use when the user asks to整理/提炼 Codex 对话, process `.codex/archived_sessions` or `.codex/sessions` rollouts, generate session packets, review packet coverage, or mark Codex sessions as distilled.
从 session-distill 生成的 packet 中导出结构化 draft memory entries。 当用户说"把 packet 变成 memory draft"、"导出记忆候选"、"为 Codex-mem 准备同步候选"时使用。
Distill Antigravity CLI sessions into review packets, local session notes, and reusable memory candidates. Use when the user asks to整理/提炼 Antigravity 对话, process `~/.gemini/antigravity-cli/history.jsonl` and brain transcripts, generate session packets, review packet coverage, or mark sessions as distilled.
Distill Grok CLI session JSONL files into review packets, local session notes, and reusable memory candidates. Use when the user asks to整理/提炼 Grok 对话, runs /session-distill, processes `~/.grok/sessions` chat history, generates session packets, reviews packet coverage, or marks Grok sessions as distilled.
Distill Hermes Agent SQLite sessions into review packets, local session notes, and reusable memory candidates. Use when the user asks to整理/提炼 Hermes 对话, process `%LOCALAPPDATA%\hermes\state.db`, generate session packets, review packet coverage, or mark Hermes sessions as distilled.
Distill OpenCode storage sessions into review packets, local session notes, and reusable memory candidates. Use when the user asks to整理/提炼 OpenCode 对话, process `~/.local/share/opencode/storage/`, generate session packets, review packet coverage, or mark OpenCode sessions as distilled.
| name | session-distill |
| description | Cursor Composer 会话 Deep Distill:packet → answer-me 验证 → session note → KB。 处理范式与 Grok 一致;禁止无验证的批量晋升。 |
| argument-hint | [status | run --next N | deep-distill | mark <id> distilled] |
Reads Cursor SQLite + JSONL agent transcripts. Default paradigm: Grok Deep Distill (3 sessions/batch).
单一真源: 本目录 ~/.cursor/skills/session-distill/ 是 Cursor 蒸馏工具的唯一安装位置。Claude Code / Codex 需要跑 Cursor 蒸馏时,也用这里的 bin/ 脚本,不要维护 ~/.claude/skills/manhua/cursor-session-distill/ 副本。
开始蒸馏前: 先过 tooling-gate.md(sync-distill-installs.py + 测试全绿)。工具没就绪不跑批量蒸馏。
~/.cursor/projects/<project>/agent-transcripts/<id>/<id>.jsonl~/.cursor/session-distill/ (CURSOR_DISTILL_DIR)E:/project/servers/.cursor/notes/conversations/session-knowledge-base.md$env:CURSOR_DISTILL_DIR = "$env:USERPROFILE\.cursor\session-distill"
# 1. Queue status
python .cursor/skills/session-distill/bin/cursor-session-distill.py list --project servers
# 2. Deep batch (3 sessions): bundle + extract claims → answer-packets
python .cursor/skills/session-distill/bin/deep-distill-run.py --offset 0 --batch-size 3
# 3. AI: answer-me verify each Q (Grep/Read/git/Shell) → fill answer-packet Results
# 4. Write session note + promote ANSWERED only → session-knowledge-base.md
# 5. check-work report → mark distilled
python .cursor/skills/session-distill/bin/cursor-session-distill.py mark <session-id> distilled
# 6. After KB promotion verified, cleanup local artifacts
python ~/.cursor/skills/session-distill/bin/cleanup-cursor-distill.py all --project servers --keep <active-session-id>
references/deep-distill-workflow.md — canonical pipelinereferences/distillation-rules.md — promotion gatesreferences/output-layout.md — paths and statusesCoverage: partial.python .cursor/skills/session-distill/bin/cursor-session-distill.py status
python .cursor/skills/session-distill/bin/cursor-session-distill.py run --next 1
python .cursor/skills/session-distill/bin/cursor-session-distill.py mark <id> distilled
python ~/.cursor/skills/session-distill/bin/cleanup-cursor-distill.py all --project servers --keep <id>