菜单
Build and validate task DAGs for Ralph parallel execution. Use when planning execution order, detecting cycles, or explaining why tasks are blocked. Triggers on: plan dag, check dependencies, why is task blocked, execution order.
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。
Generate a Product Requirements Document (PRD) for a new feature. Use when planning a feature, starting a new project, or when asked to create a PRD. The PRD should be detailed enough for execution by AI agents across tools (Amp, Cursor, Codex, etc.). Triggers on: create a prd, write prd for, plan this feature, requirements for, spec out.
Convert PRDs to prd.json format for the Ralph autonomous agent system. Use when you have an existing PRD and need to convert it to Ralph's JSON format. Ralph may run via Amp, Cursor, Codex, or similar tools, so keep output tool-agnostic unless required. Triggers on: convert this prd, turn this into ralph format, create prd.json from this, ralph json.
Merge task branches and resolve conflicts for Ralph parallel execution. Use when merging completed task branches or resolving git conflicts. Triggers on: merge branches, resolve conflict, integration merge.
Guidelines for implementing tasks safely in parallel execution. Use when working on a task in Ralph parallel mode to avoid conflicts. Triggers on: implement task safely, parallel mode, avoid conflicts.
Review integrated code for semantic and design issues. Use after merging task branches to detect inconsistencies. Triggers on: review integration, semantic review, check design consistency.
Generate and validate task metadata for Ralph parallel execution. Use when creating tasks.yaml v1 files, validating task dependencies, or checking mutex usage. Triggers on: create task metadata, validate tasks yaml, check task dependencies.
基于 SOC 职业分类
| name | dag-planner |
| description | Build and validate task DAGs for Ralph parallel execution. Use when planning execution order, detecting cycles, or explaining why tasks are blocked. Triggers on: plan dag, check dependencies, why is task blocked, execution order. |
Build, validate, and explain task dependency graphs for parallel execution.
Do NOT: author metadata, implement tasks, merge code, or review design.
Use depth-first search with three states:
unvisited - Not yet processedvisiting - Currently in recursion stackvisited - Fully processedIf we encounter a visiting node, we have a cycle.
Cycle output format:
Cycle detected: US-002 → US-003 → US-004 → US-002
A task is ready when:
dependsOn are donemutex are currently lockedWhen a task cannot run, explain why:
Task US-003 blocked:
- Waiting for: US-001, US-002 (not completed)
- Mutex locked: db-migrations (held by US-005)
Execution order (topological):
1. US-001 (no deps)
2. US-002 (no deps)
3. US-003 (after: US-001)
4. US-004 (after: US-002, US-003)
Ready queue (3 tasks can run now):
- US-001: mutex []
- US-002: mutex []
- US-005: mutex [lockfile]
Blocked (2 tasks waiting):
- US-003: waiting for US-001
- US-004: waiting for US-002, US-003
Mutex usage plan:
db-migrations: US-001, US-006 (sequential)
lockfile: US-005 (exclusive)
contract:auth-api: US-003, US-004 (sequential after US-001)
Deadlock occurs when:
Deadlock output:
DEADLOCK: No progress possible
Pending tasks: US-003, US-004
All blocked by unmet dependencies or locked mutex
US-003 needs: US-001 (failed)
US-004 needs: US-003 (blocked)
Input tasks.yaml:
tasks:
- id: A
dependsOn: []
mutex: [db-migrations]
- id: B
dependsOn: []
mutex: []
- id: C
dependsOn: [A]
mutex: [db-migrations]
- id: D
dependsOn: [B, C]
mutex: []
Output:
DAG Analysis:
Total tasks: 4
Max parallelism: 2 (A and B can run together)
Critical path: A → C → D (3 steps)
Mutex contention:
db-migrations: A, C must run sequentially
Recommended execution waves:
Wave 1: A, B (parallel)
Wave 2: C (after A)
Wave 3: D (after B, C)