菜单
Manage task memories using TCREI framework for tracking progress, decisions, and lessons. Use when planning tasks, starting work, conducting searches, receiving user feedback, or completing todos. Triggers on task lifecycle events.
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。
基于 SOC 职业分类
Use sequential thinking to reason through code implementations before writing. Invoke after planning but before editing code. References core documents in /docs/agents/ for project alignment. Triggers on "implement feature", "refactor", "architectural decision", or complex code changes.
You are operating as a Recursive Language Model (RLM). Bootstrap this skill at the start of every chat to activate a persistent Python REPL session. Use it throughout the session to explore, slice, and reason over the codebase programmatically — the way the RLM paper intended. Variables and computed state survive across every execute_code call, giving you a stateful workspace that compounds understanding across the entire conversation. Always prefer this over re-reading files or re-explaining context you have already loaded.
| name | memory-system |
| description | Manage task memories using TCREI framework for tracking progress, decisions, and lessons. Use when planning tasks, starting work, conducting searches, receiving user feedback, or completing todos. Triggers on task lifecycle events. |
Track every task with structured memories for continuity, accountability, and learning.
Execute this workflow when:
Check for existing task memory. If none exists, create one using the template in assets/memory_template.md.
| Trigger | Sections to Update |
|---|---|
| Planning | Task, Context, Rules, Evaluation |
| Search | Analysis, Feedback, Lessons |
| Task Start | Status → in_progress, branch info, commit hash |
| User Response | Analysis, Feedback, Lessons |
| Completion | Status → completed, PR URL |
Use MCP Docker tools to persist memory updates.
Structure every task memory with:
docs/ folder for background and constraintsdocs/agents/ folderUse MCP Docker knowledge graph tools:
| Operation | Tool |
|---|---|
| Create entities | create_entities |
| Update entities | add_observations, create_relations |
| Delete entities | delete_entities, delete_observations, delete_relations |
| Search/Inspect | search_nodes, open_nodes, read_graph |
Every task must:
docs/agents/Implementation Plan.mdpending | in_progress | completed | cancelled