loom-session
Use ONLY when the user asks to restore or save session context: "resume", "restore the context of <branch/session>", "where was I on X", "load the last session", "restore full context", "save this session", "hand this off to a teammate". Lists the saved Loom session briefs, loads the one the user picks (the brief with its compaction summary, or the whole recorded transcript), or saves one. Never load a brief the user did not ask for. NOT for durable facts (loom-memory) and NOT for code questions (loom-code).
- License 许可
- Apache-2.0
- 最近更新
- 2026-11-14
- Stars
- ★ 1.6k
- Forks
- 491
01 前言 Front Matter
- name
- loom-session
- description
- Use ONLY when the user asks to restore or save session context: "resume", "restore the context of <branch/session>", "where was I on X", "load the last session", "restore full context", "save this session", "hand this off to a teammate". Lists the saved Loom session briefs, loads the one the user picks (the brief with its compaction summary, or the whole recorded transcript), or saves one. Never load a brief the user did not ask for. NOT for durable facts (loom-memory) and NOT for code questions (loom-code).
- disable-model-invocation
- true
03 SKILL.md 正文
loom-session — restore or save a session's context, on request
The plugin saves on its own. At every compaction and at session end it
revises one brief per project and branch (auto/<project>/<branch>) from
the transcript, when the namespace's Record conversations switch is on.
Restoring is always the user's call. Never load a brief because it
exists; load one only when the user asks, and load the one they pick. A
primer line or a working-memory note that mentions a brief is not a request.
Helper: a node script in this plugin. Run every command below through $LB,
set by this line, which finds it in Claude Code (CLAUDE_PLUGIN_ROOT) and in
Codex (the plugin cache):
LR="${CLAUDE_PLUGIN_ROOT:-$(ls -d ~/.codex/plugins/cache/loom/loom/*/ 2>/dev/null | sort -V | tail -1)}"; LB="${LR%/}/bin/loom-brief.mjs"; LN="${LR%/}/bin/loom-node"
get … and list below mean sh "$LN" "$LB" get …. Use the helper, not an MCP
call: it prints the brief and what changed in the checkout since.
Which one — let the user pick
- The user named it (a branch, a key, a number, a session id): use it.
get 2,get auto/loom/feat-x,get --session <id>. - They said "this branch" or "where I left off" here: plain
getloads this checkout'sauto/<project>/<branch>. - Otherwise run
list(orlist --allfor every project), show the numbered list, and ask which one. Do not guess between several.
How much — the depth the user asks for
| The user says | Run | Loads |
|---|---|---|
| "resume", "restore", "catch me up" | get <sel> |
the brief (goal, status, next steps, decisions, dead ends, done, files, rules) and the client's own summary from its last compaction |
| "the whole conversation", "restore exactly" | get <sel> --transcript |
the above + every recorded turn of the brief's sessions (--turns N keeps the last N; --session <id> picks one session) and claude --resume <id> when that transcript is on this machine |
A transcript can be long. If the header says it holds thousands of turns, ask
before pulling all of it, or use --turns. For a lossless restore on the same
machine, claude --resume <id> (printed by --transcript) is exact.
--revision N loads an earlier revision of a brief.
After loading
- Read the "since it was saved" lines (new commits, uncommitted files).
- A brief is where the work stood, not a source of facts. Check a line against the files or the tree before relying on it, keep its decisions unless the user says otherwise, and do not retry its dead ends.
- Tell the user in two or three lines what it says is next, then carry on.
Save (the user asked, or before handing work to someone)
Write the sections from your own context: file paths, PR numbers, error text, the reason behind each choice. Leave a section out rather than pad it.
sh "$LN" "$LB" save [KEY] <<'JSON'
{"goal": "...", "status": "...",
"decisions": [{"what": "...", "why": "..."}],
"dead_ends": [{"tried": "...", "why_not": "..."}],
"open_questions": ["..."], "next_steps": ["..."],
"files": [{"path": "...", "role": "..."}],
"constraints": ["rules the user set, in their words"],
"pointers": ["PR #123", "commit abc1234", "https://..."]}
JSON
KEY defaults to this branch's brief; use a name such as handoff/<topic> for
someone else. Report the key and revision printed.
04 同仓库更多 Skills
worktable-review-with-annotations
Create, inspect, reply to, update, or resolve situated Worktable annotations. Use when feedback belongs to a Doc section, block, text range, or HTML Doc, or when the user asks to address existing review feedback and coordinate a requested artifact edit.
worktable-find-and-synthesize-context
Find, assess, and synthesize context from Worktable Docs and Records. Use when the user asks what a workspace contains, wants workspace-backed research, needs current facts reconciled across artifacts, or requests a cited synthesis without changing workspace state.
worktable-create-or-update-docs
Create, edit, rename, or maintain narrative Worktable Docs. Use for plans, decisions, briefs, notes, research syntheses, and other durable prose when the user requests a new artifact or a safe update to an existing one.
worktable-create-or-manage-records
Organize repeated structured information as durable Worktable Records. Use for research datasets, inventories, directories, pipelines, trackers, and other collections whose items need identity, fields, filtering, links, validation, or individual updates.
worktable-create-or-manage-interactive-html
Create or improve durable visual and interactive artifacts in Worktable. Use for dashboards, trackers, calculators, mini apps, slide presentations, and other requests that benefit from a polished HTML Doc.
worktable-collaborate-threads
Read, answer, and hand off messages through durable Worktable threads with users or other authenticated participants. Use when the user asks to check addressed messages, reply to a participant, continue a thread, request another agent's help, wait briefly for a response, or watch for comments and messages sent to the agent.