菜单
Generate accurate loom AI session digest TSVs from real question-and-answer transcripts and write them back. Use when the user asks to summarize sessions, improve vague AI conversation titles, run loom session gen/set/ls, batch process Claude sessions, or make answer-side content searchable.
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。
确保 loom CLI 已安装并在 PATH 上。当准备使用 loom 但命令行里 `loom` 不存在(command not found / 首次使用 loom 插件)时,先用本技能一行装好底层 CLI,再继续。仅负责安装引导,不负责日常使用(用法见 loom 技能)。
同步/检索个人跨工具成果台账 loom。当用户想「记录今天做了什么」「同步 loom」「查我以前/某天/某件事做过什么、结论是什么、相关文档/数据在哪」「生成日报/周报/绩效素材」时使用;先用它查历史,别只翻当前仓。
Generate and write back AI-authored loom daily reports from real collected evidence. Use when the user asks for a daily report, work log, day summary, report gen/set workflow, or wants to sync loom, draft 今日工作/今日思考/明日计划, and store the result.
Guide a new or returning user through loom setup, configuration, first sync, history intake, and the code-vs-data boundary. Use when the user asks to initialize loom, follow ONBOARDING.md, configure sources, collect historical work traces, or explain what lives in the repo versus ~/.loom.
Classify loom entries into topic leaves with a gather-to-TSV-to-apply workflow and post-apply verification. Use when the user asks to triage topics, build or refine the topic DAG, run loom topic gather/apply/show, review topic assignments, or classify historical entries without keyword overreach.
Inspect loom vault backup safety before git push or cloud backup. Use when the user asks to check vault state, gitignore coverage, remote configuration, tracked raw data, push risk, private backup readiness, or whether loom sync --push is safe.
| name | loom-session-digest |
| description | Generate accurate loom AI session digest TSVs from real question-and-answer transcripts and write them back. Use when the user asks to summarize sessions, improve vague AI conversation titles, run loom session gen/set/ls, batch process Claude sessions, or make answer-side content searchable. |
This workflow currently summarizes Claude sessions with readable local transcripts. Cursor and Codex entries may not have message-level text available, so do not promise full transcript digests for them.
Session digests are derived sidecar data in ~/.loom/data/session_digests.json. They are applied back onto entries after sync, so regenerated collection does not erase them.
Refresh first unless the user asks not to:
loom sync
loom session gen YYYY-MM-DD
If the output says there are no summarizable Claude sessions, stop and report that fact.
For each SESSION <id> block, write exactly one TSV row:
session_id<TAB>准确标题<TAB>2-4句可检索摘要
Rules:
loom session set rejects IDs that are not real sessions for that date.ref to inspect the raw transcript for that date before writing.Save the TSV locally, then run:
loom session set YYYY-MM-DD --file <digest.tsv>
Use --push only after explicit user confirmation.
Run:
loom session ls
loom search <distinct-digest-keyword>
Confirm that titles are updated, the digest text is searchable, and the journal now shows improved session titles.
For bulk history work, process by date in small batches and verify after each batch before moving on.