Agent Skills for TiMEM memory workflows. Skills orchestrate when and how to call timem-mcp atomic tools — memories (search_memories, create_memory, delete_memory) and rules (recall_rules, learn_rule, record_rule_outcome).
Language: English | 简体中文
Skills follow the Agent Skills open standard (portable across Cursor, Claude Code, Codex, and other compatible agents).
If TiMEM MCP is already connected and you write code, install one coding package:
| Package | Path | When |
|---|---|---|
| Full (progressive disclosure) | dist/full/timem-coding-memory/ |
Default — multi-file skill |
Standalone (single SKILL.md) |
dist/standalone/timem-coding-memory/ |
Simplest copy — one markdown file |
Copy the folder into your client's skills directory (name must stay timem-coding-memory):
| Client | Project | User (global) |
|---|---|---|
| Agents / Codex-style | .agents/skills/ |
~/.agents/skills/ |
| Claude Code | .claude/skills/ |
~/.claude/skills/ |
| Cursor | .cursor/skills/ |
~/.cursor/skills/ |
git clone https://github.com/TiMEM-AI/timem-skill.git
cd your-project
# Full package (recommended)
mkdir -p .agents/skills # or .claude/skills / .cursor/skills
cp -r /path/to/timem-skill/dist/full/timem-coding-memory .agents/skills/
# Or standalone (single SKILL.md inside the folder)
cp -r /path/to/timem-skill/dist/standalone/timem-coding-memory .agents/skills/Helper:
python /path/to/timem-skill/scripts/install.py --skill coding --target agents
# python .../install.py --skill coding-standalone --target cursor --globalNo separate shared folder is required for install. Packages are self-contained.
| Skill | Scene | Install from |
|---|---|---|
| timem-coding-memory | coding |
dist/full/ or dist/standalone/ (prefer these) |
| timem-general-memory | general |
dist/full/ or dist/standalone/ (prefer these) |
| timem-writing-memory | writing |
skills/timem-writing-memory/ (self-contained) |
| timem-rule-learning | rules (cross-scene) | skills/timem-rule-learning/ (self-contained) |
| timem-knowledge | knowledge | skills/timem-knowledge/ |
Each skill is one job. The agent routes by each skill's description. timem-rule-learning is not a memory scene: it runs alongside memory skills and is scoped by agent_id.
Source of truth for development lives under skills/. User-facing coding and general packages are built into dist/ via python scripts/build-all.py.
- Configure timem-mcp in your MCP client (
TiMEM_API_KEY,TiMEM_USER_ID). - Install the skill package(s) you need (see above and docs/installation.md).
Skill = orchestration (search/write memories; recall/learn/grade rules). MCP = atomic API tools.
Details: docs/architecture.md
MIT — see LICENSE.