Tasks

Skill MakFly/ai-dev-kit/.claude/skills/tasks

Spec-Driven Agentic OS for Claude Code + OpenAI Codex CLI. Native skills, subagents, hooks, and an adversarial codex review loop. Inspired by github/spec-kit, built on Claude Code primitives. No daemon, no vendor lock-in.

Install

npx -y skills add MakFly/ai-dev-kit --skill tasks

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

2 stars2 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Use PROACTIVELY when the user invokes /tasks <slug> or asks "break this plan into tasks". Runs the `decomposer` subagent in a forked context which writes specs/<slug>/tasks.md. Refuses if plan.md is still `draft`.

SKILL.md

1.6 KB, as published. Nobody here has run it

/tasks

Author specs/<slug>/tasks.md via the decomposer subagent.

Procedure

Read plan. cat specs/<slug>/plan.md. Refuse if missing or Status: draft (plan must be reviewed first).
Cross-check acceptance mapping. Plan §8 must already map spec criteria to plan elements. If missing, send back to /plan.
Delegate. Invoke decomposer with the slug.
Print summary. Count of tasks, total estimated time, list of T-IDs with one-line title.
Validate coverage. Every spec §4 criterion appears in tasks.md acceptance-mapping table — refuse if not.
Commit. git commit -m "tasks(<slug>): decompose plan into T1..TN".

Doctrine

Atomic > comprehensive. A task that takes > 1 hour is too big — split.
Owner is named. No "team", no "claude". One agent or user per task.
Acceptance is testable. Each task's acceptance line must be a runnable command or a grep-able assertion.
Coverage is enforced. No spec §4 criterion may be unmapped.

On completion

Suggest: /implement <slug> to begin execution.