What problem does it solve? Turning a feature request or bug fix into reviewed, working code usually requires ad-hoc planning, manual code review, and untracked scope creep. This Skill structures that entire cycle: it clarifies intent, writes a bounded specification, implements the change, runs multi-agent adversarial review, and produces a human-reviewable diff trail. ## Core Features & Use Cases - Spec-driven development: Generates a token-bounded spec (900–1600 tokens) with intent, boundaries, I/O edge-case matrix, tasks, and acceptance criteria before any code is written. - Dual routing: Chooses between a one-shot path for zero-blast-radius changes and a full plan-code-review path for everything else. - Adversarial review loop: Launches blind, edge-case, and acceptance-auditor review subagents, classifies findings (intent_gap, bad_spec, patch, defer, reject), and loops back to re-derive code when needed. - Epic context compilation: Distills PRD, architecture, and UX planning documents into a focused per-epic context file, and syncs story status with sprint-status.yaml. - Use Case: Ask it to "add pagination to the orders API" and it will investigate the codebase, draft a spec for your approval, implement the change, run three independent reviews, fix findings, commit, and open a clickable Suggested Review Order in your editor. ## Quick Start Ask the agent to implement a feature or fix a bug, for example: implement user session timeout handling following the project's existing patterns.
