What problem does it solve? Delegating implementation work to a fast Executor agent fails when task specs are vague, rest on unverified assumptions, or receive uniform verification regardless of risk. This Skill provides rules for writing specs that are hard to misimplement and for matching review depth to actual defect risk. ## Core Features & Use Cases - Spec authoring rules: Enforce narrow task granularity, the negative-claim rule (mandating full regression runs when correctness depends on something being unused or unreachable), caller tracing across files, grounding in real code with re-grep instructions, and one-task-one-commit discipline. - Cross-engine spec review: Submit draft specs to independent AI engines or blind self-review instances in iterative cycles before finalizing, keeping at least one unprimed reviewer per cycle. - Parallel batch coordination: Define disjoint write scopes, constrain shared barrel files, pre-assign shared type names, and pin toolchain versions to make merges cheap. - Risk-calibrated verification: Scale review from a simple diff plus smoke test for narrow tasks up to full regression sweeps for negative-claim tasks, always against disposable data copies. - Use Case: Before handing a multi-file refactoring task to an Executor, write a spec citing exact functions with re-grep warnings, trace all callers of shared helpers, and require a full test suite run in the Verification section. ## Quick Start Use the task-spec-authoring skill to write an Executor task spec for converting the authentication module, including caller tracing and a risk-appropriate verification section.
