一键导入
task-manager
Persistent task management with subagent orchestration. Create tasks, assign to agents, track execution, and build workflows via js-doc-store-server.
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。
菜单
Persistent task management with subagent orchestration. Create tasks, assign to agents, track execution, and build workflows via js-doc-store-server.
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
| name | task-manager |
| version | 1.0.0 |
| tags | tasks,workflow,subagent,project-management,orchestration,crud |
| author | MauricioPerera |
| description | Persistent task management with subagent orchestration. Create tasks, assign to agents, track execution, and build workflows via js-doc-store-server. |
This skill introduces a task management system backed by js-doc-store-server that integrates with the subagent framework. Tasks are persistent objects that can be assigned to specific agents, executed in isolation, and tracked across sessions.
| Problem | Solution |
|---|---|
| Ad-hoc subagent calls get lost | Tasks persist in tasks table |
| No visibility of what was delegated | Task list shows status, agent, result |
| No retry mechanism | Failed tasks can be re-executed |
| No batching | Parent/child tasks for complex workflows |
| No cross-session continuity | Server storage survives pi restarts |
Table: tasks
| Field | Type | Description |
|---|---|---|
id | text | Unique task identifier |
title | text | Short name |
description | text | Detailed instructions |
status | text | pending / in_progress / completed / failed / cancelled |
priority | text | low / medium / high / critical |
agent | text | Subagent name (scout, planner, worker, etc.) |
agent_scope | text | user / project / both |
input | text | Context/data passed to agent |
output | text | Result from agent execution |
tags | text | Comma-separated labels |
parent_id | text | For subtasks |
conversation_id | text | Link to persistent messages |
error_message | text | If execution failed |
attempts | number | Execution retries |
created_at | text | ISO timestamp |
started_at | text | ISO timestamp |
completed_at | text | ISO timestamp |
task_managertask_manager({
action: "create", // create | list | get | update | assign | execute | delete
title: "Add auth middleware",
description: "Implement JWT validation...",
priority: "high",
agent: "worker",
input: "Current auth.ts is empty, see repo structure...",
tags: "auth,security,sprint-3",
parent_id: "task-parent-001"
})
| Action | Required params | Description |
|---|---|---|
create | title | Creates pending task, returns ID |
list | — | Lists tasks with optional filters |
get | id | Full task details |
update | id + fields | Modify title, status, output, etc. |
assign | id + agent | Bind agent, validates agent exists |
execute | id | Spawns subagent, saves result, updates status |
delete | id | Remove task |
task_manager({ action: "execute", id: "task-xxx" })
│
├──► 1. Load task from server
│ ├──► If no agent assigned → error
│ └──► If task not found → error
│
├──► 2. Resolve agent definition
│ ├──► Try server table "agents"
│ └──► Fallback to ~/.pi/agent/agents/*.md
│
├──► 3. Spawn pi --mode json
│ ├──► Agent markdown as system prompt
│ └──► Task description + input as user prompt
│
├──► 4. Capture JSONL output
│ └──► Extract assistant text block
│
└──► 5. Update task in server
├──► success → status: completed, output: text
└──► failure → status: failed, error_message: stderr
// 1. Create
const t = await task_manager({
action: "create",
title: "Refactor database connection",
description: "Move DB config to environment variables",
priority: "high",
tags: "refactor,database"
});
// 2. Assign
await task_manager({ action: "assign", id: t.id, agent: "worker" });
// 3. Execute
await task_manager({ action: "execute", id: t.id });
// 4. Review result
await task_manager({ action: "get", id: t.id });
// Parent: plan
const plan = await task_manager({ action: "create", title: "Plan OAuth integration", agent: "planner" });
await task_manager({ action: "execute", id: plan.id });
const planResult = await task_manager({ action: "get", id: plan.id });
// Child: implement based on plan output
const impl = await task_manager({
action: "create",
title: "Implement OAuth",
input: planResult.output,
parent_id: plan.id,
agent: "worker"
});
await task_manager({ action: "execute", id: impl.id });
// List all pending auth tasks
await task_manager({ action: "list", status: "pending", tags: "auth" });
// List all tasks assigned to worker
await task_manager({ action: "list", agent: "worker" });
/taskQuick UI without tool calls:
/task list # Show last 20 tasks
/task create Fix login # Create task with title
| Component | Role |
|---|---|
subagent tool | Direct ad-hoc delegation (no persistence) |
task_manager tool | Persistent, trackable, retryable delegation |
agents table | Agent definitions (shared with subagent discovery) |
messages table | Conversation logs linked via conversation_id |
inputattempts before retry: Don't loop foreverskill-registry — Dynamic skill discoveryagent-registry — Hybrid agent discoverymemory-management — Persist conversationssubagent extension — Direct delegation toolPi extension for workflow automation enabling resilient server startup procedures and agent-based workflow orchestration.
Meta-skill pattern for dynamic skill discovery. Keep only ONE SKILL.md on filesystem; query all specialized skills on-demand from js-doc-store-server by tags.
Dynamic agent registry via js-doc-store-server. Store, version, and discover subagent definitions in a database instead of static markdown files.
Professional Data Architect methodology. Includes entity analysis, schema design, implementation workflow, and prebuilt patterns (CRM, Wiki, Inventory).
Persist Pi conversation messages to js-doc-store-server. Auto-incremented turns, recovery after compaction, admin queries, best practices. v2.2.1 with 4 core tools.
RAG without embeddings using hierarchical Reasoning Trees. Navigate Root→Branch→Leaf, maintain summaries, and retrieve via structure rather than vectors.