What problem does it solve? When an AI agent implements tasks from a spec, it can drift: inventing scope, silently fixing other domains, closing tasks with failing checks, or leaving untracked temporary hacks. This Skill defines the rules of conduct for executing spec tasks so deliveries stay coherent, auditable, and within the declared domain boundary. ## Core Features & Use Cases - Spec-driven task execution: Locates tasks by stable identifiers in tasks.md, reads spec artifacts and authoritative domain documentation, and delivers only the delta between current code and the target state. - Quality-checks policy: Runs only the checks the project declares, allows justified skipped results, limits fix attempts to three, and separates pre-existing failures from introduced ones. - Forward-dependency tracking: Records deliberate temporary constructs in forward-deps.md with code markers, and requires destination tasks to resolve them before closing. - Escalation discipline: Defines exactly when to stop and ask the human — drift between documentation and code, dead code, scope gaps, ungrounded dependencies — instead of guessing. - Use Case: An agent receives tasks T3 and T5 from .agents/specs/012-reporting/. It reads the spec, implements only in-scope changes, runs the project's lint and tests, reports impacted test cases from test-cases.md, and marks the checkboxes only when work is genuinely complete. ## Quick Start Ask the agent to implement tasks T3 and T5 from the spec in .agents/specs/ following the task-execution rules.
