一键在 Manus 中运行任何 Skill

store

星标0

分支0

更新时间2026年7月7日 13:32

This skill should be used when the user asks for maintenance or analysis across the WHOLE memory store — "audit the whole store", "store-wide retro", "check every project for X", "rebuild all the indexes", "run an agent inside the memory dir" — or invokes /tm:store <task>. Store-root scope (all projects plus global), unlike the project-scoped verbs. Dispatches a delegate agent inside the memory dir: read-then-report first, then every proposed write is confirm-gated per action.

安装

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。

在 Manus 中运行

来源

Lucklyric

Lucklyric/turing-mem

打开 GitHub 仓库查看创作者相关仓库

下载

在 Manus 中运行

tm — store-level maintenance (root scope)

Runs a delegated agent inside the memory dir, across all projects/* and global/ — for system/store-level work the project verbs don't cover. Read-then-report first; nothing is written without per-action confirmation.

Guard (run first, always)

Run ${CLAUDE_PLUGIN_ROOT}/scripts/tm-guard.sh "$PWD" and read its KEY=VALUE output. Require TM_STATUS=OK — otherwise stop and follow the repair flows in <MEMORY_DIR>/references/bootstrap.md. PROJECT_EXISTS is irrelevant here: this verb is store-scoped, not project-scoped. Keep MEMORY_DIR.

Steps

Pin the task. Restate the user's task in one line. If no task was given, ask for one (examples: index-drift audit across all projects, store-wide retro, orphaned-notes sweep, ## Projects table reconciliation).
Dispatch tm-archivist via the Task tool — report pass. Pass exactly:
- mode: store
- memory_dir: <MEMORY_DIR>
- project_path: .
- project_id: store
- task: <the pinned task> It investigates the whole store and returns a numbered plan (≤10 lines), writing nothing.
Relay the plan and confirm via AskUserQuestion which actions to apply (multi-select: each action / all / none). Purely informational tasks end here — relay the findings and stop.
Apply pass. Re-dispatch tm-archivist with the same inputs plus apply: <confirmed numbers>. It executes exactly those actions. Relay its per-action report. The store git rule applies (own-repo commits itself; nested stores are committed by the parent repo's workflow).

Scope map (which verb owns what)

Project scope: remember, consolidate, retro, reindex (default), link.
Project + global read: recall.
Store root: store (this verb), reindex global, init (setup/repair), status (reports both).
Project → global: promote (the only path that writes global memory content).

Codex fallback

Not installed on Codex — run store-level maintenance from Claude Code. For a read-only look on Codex, follow the retrieval protocol in the store's AGENTS.md instead (never bulk-load the store).

同仓库更多 Skills

同仓库

server

Lucklyric/turing-mem

This skill should be used when the user asks to "start the memory server", "open the memory browser", "visualize my memory", "stop the tm server", "is the memory server running", or invokes /tm:server [start|stop|status]. Manages a local read-only HTTP browser for the whole store - global and project memory, paginated lists, keyword search - at http://127.0.0.1:<port>/. Starting is confirmed with the port and store path; stop and status act on a clear ask.

2026-07-150

checkup

Lucklyric/turing-mem

This skill should be used when the user asks to "run a memory checkup", "check up on my memory", "take care of my memory", "tidy up memory", "do whatever memory needs", "what should memory do next", or invokes /tm:checkup. Auto triage across the whole turing-mem surface - one status read, then a priority ladder picks the single best next action (repair, link, doc refresh, consolidate, checkpoint, commit reminder) and dispatches the matching verb. When the project is unresolved it searches the store for candidate folders and presents concrete choices. Every action goes through that verb's own confirm gates; nothing is decided silently.

2026-07-130

init

Lucklyric/turing-mem

This skill should be used when the user asks to "set up turing-mem", "initialize the memory store", "init tm", "repair the memory store", accepts an offer to set up after a guard reports a non-OK status, or invokes /tm:init. Creates config.json and the memory dir, adopts an existing store, scaffolds the current project folder, and walks the bootstrap repair flow. Every significant step is confirmed with the user before acting — nothing is created without explicit approval.

2026-07-130

link

Lucklyric/turing-mem

This skill should be used when the user asks to "link this folder/project to a memory project", "map this directory to project X", "pin this repo to existing memory", accepts a KEY_MISMATCH repair, or invokes /tm:link [id|path]. Forces the current directory to map to an existing turing-mem project folder by writing project_overrides in config.json — cross-host linking for no-remote projects. The mapping is always confirmed before writing.

2026-07-130

rename

Lucklyric/turing-mem

This skill should be used when the user asks to "rename the project memory folder", "change the memory folder name", "rename project X in memory", "give this project's memory a different name", or invokes /tm:rename [new-name]. Renames a project's store folder prefix — the -8hex identity suffix is always preserved, so resolution keeps working from every clone and host. Updates the global index row and any config overrides pointing at the folder. The rename is always confirmed before anything moves.

2026-07-080

status

Lucklyric/turing-mem

This skill should be used when the user asks to "check turing-mem status", "tm status", "memory status", "is my memory store healthy", "tm health", or invokes /tm:status. Read-only health report — config, store version, project resolution, note and checkpoint counts, consolidation state — with the matching repair verb for anything unhealthy.

2026-07-080

name

store

description

tm — store-level maintenance (root scope)

Guard (run first, always)

Steps

Pin the task. Restate the user's task in one line. If no task was given, ask for one (examples: index-drift audit across all projects, store-wide retro, orphaned-notes sweep, ## Projects table reconciliation).
Dispatch tm-archivist via the Task tool — report pass. Pass exactly:
- mode: store
- memory_dir: <MEMORY_DIR>
- project_path: .
- project_id: store
- task: <the pinned task> It investigates the whole store and returns a numbered plan (≤10 lines), writing nothing.
Relay the plan and confirm via AskUserQuestion which actions to apply (multi-select: each action / all / none). Purely informational tasks end here — relay the findings and stop.
Apply pass. Re-dispatch tm-archivist with the same inputs plus apply: <confirmed numbers>. It executes exactly those actions. Relay its per-action report. The store git rule applies (own-repo commits itself; nested stores are committed by the parent repo's workflow).

Scope map (which verb owns what)

Project scope: remember, consolidate, retro, reindex (default), link.
Project + global read: recall.
Store root: store (this verb), reindex global, init (setup/repair), status (reports both).
Project → global: promote (the only path that writes global memory content).

Codex fallback

Not installed on Codex — run store-level maintenance from Claude Code. For a read-only look on Codex, follow the retrieval protocol in the store's AGENTS.md instead (never bulk-load the store).