What problem does it solve?
Coordinating a team of AI agents to implement, review, and merge an entire Spec's sub-issues without human supervision is error-prone: dependencies get missed, merges conflict, and interrupted runs lose state. This Skill acts as a team-lead that plans, dispatches, integrates, and adjudicates a batch of GitHub issues until every stage PR is merged or explicitly paused.
Core Features & Use Cases
- Batch planning and dependency projection: Polls a Spec's sub-issues or an explicit issue list via the GitHub CLI, builds the canonical blocked-by graph, and partitions work into minimal independently mergeable stages.
- Role-based agent delegation: Dispatches implementer, local-reviewer, and review-looper agents with contract prompts, isolated git worktrees, per-issue conventional commits, and a stage-level AI review loop that converges on a green, mergeable HEAD.
- Crash-safe ledger and recovery: Appends decision, authorization, fault, and pause events to a thin JSONL ledger under
.afk/, enabling a new team-lead to resume an unfinished batch from remote truth.
- Use Case: Point it at Spec #42 with twelve sub-issues; it plans three stages, runs implementer/reviewer relays per issue, opens staged draft PRs with
Closes references, drives each PR through AI review to green, rebase-merges them in order, and posts a human QA checklist on the Spec.
Quick Start
Ask the agent to use the afk-team-workflow skill to run Spec #42's sub-issues unattended until all stage PRs are merged.