--- name: synthesis-adversarial-review description: "Run bounded adversarial review by differently shaped agents against the principal's outcome: artifact-complete rounds, production-topology handoffs, concessions, a fail-closed finding ledger, sufficiency rulings, post-publication acceptance. Use for adversarial, cross-agent or red-team review." license: "Apache-2.0" depends_on: ["synthesis-grounding-discipline", "synthesis-anti-shortcuts", "synthesis-project-management"] metadata: author: "Rajiv Pant" version: "2.0.0" source_repo: "github.com/synthesisengineering/synthesis-skills" source_type: "public" format: v5 --- # Synthesis Adversarial Review Differently shaped agents attack the same work from different blind spots, in bounded rounds, to deliver what the principal asked to ship. Before freezing or accepting a package, apply the mandatory [domain contracts and replay scorecard](references/domain-review-contract.md); validate exact source bindings with `scripts/review_contract.py` and retain the existing ledger/action owners. ## Binding rules The headings below are the protocol's sections, in the order `scripts/protocol_acceptance.py` checks. [references/protocol.md](references/protocol.md) holds each section in full. ### Purpose 1. **The review exists to deliver the principal's outcome:** the artifacts and enforced boundaries the principal asked to ship, at the accepted quality bar. Reviewer satisfaction, control growth and a large finding count are not completion criteria. 2. **The review grants no publication, deployment, communication or repair authority.** Apply the shared [decision ownership contract](../synthesis-thinking-framework/references/decision-ownership.md) and record the technical sufficiency owner first. 3. **For [typed domain review](../synthesis-autopilot/references/domain-quality.md), freeze the accepted requirement, source universe, domain method and rubric before the attempt.** Replacing a reviewer without changed work or a concrete open risk does not justify another round. ### Before Round One: Proportionality Contract 4. **Record the contract before dispatching a reviewer:** the principal outcome; the Closed review universe; consequence and depth; a Round-trip budget that counts principal courier crossings; and the Stop rule. ### Roles and Blind-Spot Rotation 5. **Executor and Adversarial reviewer are separate roles.** Rotate the blind spot, not the agent name. Concession is health: record what each side conceded. ### Adjudication and Separation of Duties 6. **An adjudicator settles surviving disagreements** and owns ledger transitions. The repairer never edits, moves or deletes the reviewer's reproducer without the adjudicator's recorded approval. ### Goal-Focused Round 7. **Every goal-focused round runs five terminal stages:** 1. **Contract.** Restate the outcome, decision owners, approval gates, artifact universe and attack plane. 2. **Attack.** Derive counterexamples from the production path; encode real defects as failing fixtures before repair. 3. **Disposition.** One terminal row per artifact and finding; prose and executable repair are not interchangeable. 4. **Concept sweep.** Search the whole evidence package for the claim a repair displaced. 5. **Sufficiency.** Record established, open, and risk of shipping now. At a principal-owned or supervised checkpoint, the principal's ruling terminates the loop. 8. **A `ship-blocking` finding stays in the delivery** until repaired, conceded or resolved by its decision owner; a `ship-improving` finding names its follow-up project. A repair reopens only its failed criteria. ### Sidecars, Evidence, and Handoff Topology 9. **Sidecars are claims.** Every handoff names the production entry point, the enforcing boundary, the receipt consumer, the exact artifacts, each finding's reproducer, the concept sweep owed after a provenance correction, and what the evidence does not verify. ### Finding Ledger 10. **One ledger per engagement, edited only through `scripts/finding_ledger.py`** (command below). Each finding carries a state, a classification, an authority label, a separate enforcement outcome, evidence, and a follow-up project when ship-improving. ### Bounded Control Depth 11. **A finding in generation N+1 of an unrequested control stops control growth;** generation N+2 needs an explicit principal decision. ### Bounded Post-Publication Acceptance 12. **After publication, a second agent derives the live universe from destination state** and records a per-artifact matrix. A correction beyond the current grant needs fresh approval. ### Agent-Principal Norms 13. **Surface a known-false claim once,** in the principal's terms. A proposed loosening nominates the loosener for review. Approval fatigue is a failure mode. ### Completion Report 14. **Report the principal outcome first** and name the unverified remainder. "The reviewer is satisfied" is never a completion signal. ## Contents - [references/protocol.md](references/protocol.md): every protocol section in full, with the reasons, the ledger field list and the post-publication steps. Read the matching section before running that stage of a review. - [references/domain-review-contract.md](references/domain-review-contract.md): the domain review package, its four profiles and seven records, mechanical controls and the replay scorecard. Read it before freezing or accepting any package. - [references/profiles.json](references/profiles.json) and [references/schemas/](references/schemas/): the review profiles and the seven record schemas that `scripts/review_contract.py` validates. Read them when building a review package. - [references/coverage-map.md](references/coverage-map.md): where each part of the 1.2.1 text now lives. - Finding ledger command: below. ## Finding ledger command Create one YAML ledger per engagement in the owning project's `resources/` directory: ```bash python3 scripts/finding_ledger.py init \ --resources-root resources \ --file resources/-findings.yaml \ --engagement \ --principal-outcome '' \ --round-trip-budget \ --proportionality 'AGENT HEURISTIC: ' ``` Use `validate` before every handoff.