am-will/codex-skills

Overview

This skill orchestrates parallel Sparky subagents to execute tasks from a markdown plan. It only runs when explicitly invoked with a /parallel-task-spark command and coordinates task waves, dependencies, commits, and plan updates until the plan is complete.

How this skill works

The orchestrator parses the plan file to identify task IDs, dependency graphs, full task content, and optional task subsets. It launches only sparky-role subagents for every unblocked task in parallel, validates their outputs and plan updates, and repeats waves until all tasks are completed and logged in the plan file.

When to use it

• You have a development plan in markdown with discrete tasks and dependency metadata.
• You want tasks executed in parallel where dependencies determine order.
• You need automated enforcement that each task updates and logs completion in the plan.
• You want automated retries or escalation when subagents fail to meet acceptance criteria.

Best practices

• Provide a well-structured plan file: clear task IDs, depends_on lists, acceptance criteria, and validation steps.
• When running a subset, include any required dependent task IDs to avoid missing prerequisites.
• Keep each task scope atomic so subagents can commit small, independent changes.
• Ensure plan filenames and file paths in tasks are accurate to prevent path-related blockers.
• Review execution summaries after each wave before launching the next wave.

Example use cases

• Execute a full release plan where backend, frontend, and infra tasks have cross-dependencies.
• Run a focused subset of tasks for a hotfix by specifying task IDs on the command line.
• Orchestrate parallel implementation of feature branches that share common setup tasks.
• Validate that every task writes a concise work log and commit entry into the plan markdown.