What problem does it solve? After a feature spec is written, teams must decide how large the adversarial review council should be and how deep the task workforce should run, but they lack grounded cost data to make that call. This Skill automates that decision step: it computes a recommendation from the spec's own signals, backs it with exact session counts and measured token/time evidence, and writes the accepted configuration into the feature's profile.yaml. ## Core Features & Use Cases - Signal-based recommendation: Counts functional requirements, user stories, key entities, and cited contracts in spec.md to recommend council_members (1-8), workforce_depth (full|light), and lane (small|standard), each with a stated rationale. - Grounded cost evidence: Relays exact session counts and token/time figures verbatim from cost_evidence.py, distinguishing local repository history from foreign baseline data and never estimating unavailable numbers. - Safe profile writing: Never overwrites an existing profile.yaml, validates operator overrides against schema constraints, resolves lane composition, and runs interactively or automatically depending on context. - Use Case: After running /speckit-specify for a new feature, this hook fires automatically, presents a recommendation block with cost evidence, asks once for acceptance or override, and writes profile.yaml before the plan phase begins. ## Quick Start Run /speckit-run-config after specifying a feature to get a recommended council size and workforce depth with cost evidence written into the feature's profile.yaml.
