code

name	code
description	Execute tasks from a tasks.yaml by orchestrating code-writer agents
argument-hint	<path-to-tasks-yaml> [--task <task-key>]

Execute tasks from a tasks.yaml file by orchestrating code-writer agents. Automatically handles dependency resolution, parallel execution, and status tracking.

Usage

/code <path-to-tasks-yaml> [--task <task-key>]

Parameters

• path-to-tasks-yaml: Path to tasks.yaml file (required)
• --task: Execute only a specific task key, skipping dependency resolution

Examples

/code prd/auth-feature/tasks.yaml
/code ./tasks.yaml --task setup-database
/code prd/search/tasks.yaml

Implementation

1. Load and validate tasks.yaml:

   uv run ~/.claude/skills/tasks/tasks.py <path-to-tasks-yaml> summary
   

   Report task summary:

   • Total, done, in-progress, open counts

2. If --task specified: Jump directly to step 5 for that task.

3. Find ready tasks — tasks where ALL of the following are true:

   • status == "open"
   • All keys in depends have status == "done"

   uv run ~/.claude/skills/tasks/tasks.py <path> ready
   

4. If no ready tasks:

   • All done → report completion and stop
   • Circular deps or all blocked → report which tasks are blocking and stop
   • Tasks in progress → report waiting state

5. Determine execution strategy for ready tasks:

   Parallelize (max 3 concurrent agents) when tasks touch completely separate files/domains.

   Run sequentially when tasks:

   • May modify the same file(s)
   • Share configuration or state files
   • Have overlapping file paths
   • When in doubt → sequential

6. For each ready task:

   a. Mark in progress:

   uv run ~/.claude/skills/tasks/tasks.py <path> set <task-key> progress
   

   b. Gather context:

   • Read the PRD file in the same directory (prd.md)
   • Read the task's detail file if details path is set
   • Read detail files from completed dependent tasks
   • Scan codebase for relevant existing patterns

   c. Choose agent type:

   • Single file, simple change → code-writer-simple
   • Multi-file, architectural, or complex → code-writer-complex

   d. Invoke agent using the prompter skill with this structure:

   <delegation>
     <role>Senior software engineer specializing in {detected language/framework}</role>
   
     <context>
       <prd>{summary from prd.md}</prd>
       <dependencies>{list of completed dependent tasks and what they built}</dependencies>
       <codebase>{relevant existing files, patterns, conventions}</codebase>
     </context>
   
     <task>
       <key>{task-key}</key>
       <description>{task.description}</description>
       <details>{contents of task detail file if exists}</details>
     </task>
   
     <constraints>
       - Follow existing codebase conventions
       - Ensure compatibility with completed dependent tasks
       - Write production-ready code
       - No new external dependencies without explicit approval
     </constraints>
   
     <verification>
       1. Code compiles/runs without errors
       2. Existing tests still pass
       3. Implementation matches task details
     </verification>
   
     <return>
       - Files created/modified with paths
       - Summary of implementation
       - Any issues or blockers encountered
     </return>
   </delegation>
   

   e. On success, mark done:

   uv run ~/.claude/skills/tasks/tasks.py <path> set <task-key> done
   

   For parallel tasks: Launch all agents in a single tool call block.

7. Loop: After completing a batch, go back to step 3 to find the next ready tasks.

8. Final verification when all tasks report done:

   uv run ~/.claude/skills/tasks/tasks.py <path> verify
   

Task Management Reference

Operation	Command
Summary	uv run ~/.claude/skills/tasks/tasks.py <path> summary
List all tasks	uv run ~/.claude/skills/tasks/tasks.py <path> list
View ready tasks	uv run ~/.claude/skills/tasks/tasks.py <path> ready
View in progress	uv run ~/.claude/skills/tasks/tasks.py <path> list --status progress
Mark in progress	uv run ~/.claude/skills/tasks/tasks.py <path> set KEY progress
Mark done	uv run ~/.claude/skills/tasks/tasks.py <path> set KEY done
Verify all done	uv run ~/.claude/skills/tasks/tasks.py <path> verify

Error Handling

• Task fails: Leave status as "progress", report the error, allow user to retry
• No ready tasks but not all done: Report which tasks are blocking and why
• Missing dependency: Report which upstream tasks need to complete first
• Missing detail file: Proceed using only the task description and PRD context

Output Format

✓ Loaded tasks.yaml (X tasks)
  - Done: Y
  - In Progress: Z
  - Open: W

▶ Ready: task-name-1 (no deps)
▶ Ready: task-name-2 (no deps)

[Agents executing...]

✓ task-name-1 complete
  Modified: src/foo.ts, src/bar.ts

✓ task-name-2 complete
  Created: src/baz.ts

▶ Ready: task-name-3 (depends: task-name-1, task-name-2 ✓)

[...]

✓ All tasks complete!