What problem does it solve? Running a complex coding task with AI agents often means trusting the model's own claims of success. This Skill removes that trust requirement by orchestrating a full agent pipeline where tests are authored blind, cryptographically locked, and judged by an independent verifier — so a passing result is backed by tamper-evident evidence, not model confidence. ## Core Features & Use Cases - Locked verification pipeline: Coordinates researcher, planner, architect, verifier-author, executor, verifier-judge, security-auditor, and observer agents in a fixed sequence with a bounded 3-attempt retry loop. - Tamper-evident test integrity: Hashes tests with sha256, locks them at the filesystem level, ties them to a scope-guard hook via .harness/run.json, and brackets the judge step with before/after fingerprints to detect tampering. - Workflow-aware routing: Resolves frozen workflow profiles, enforces a rigor guard (quick/mid tasks are rejected to the light path), asserts the hard backbone agents are present, and optionally fans out multi-agent author/executor batches through Workflow calls. - Use Case: A developer fills a task spec with acceptance criteria for a cross-module feature, then invokes this Skill to run the full pipeline — tests are authored blind from the spec, locked, the executor implements against them, and an independent judge returns a typed PASS/FAIL/PARTIAL verdict. ## Quick Start Run the filled task spec at path/to/task-spec.json through the full locked verification pipeline using prepare-run and the run-harness-task procedure.
