AdversarialScanner

Task

Attack a post-clarification specification for internal consistency failures and produce a ranked list of stress-test findings.

Inputs

Specification file path and adversarial stress-test heuristics.

Execution Rules

Analyze only the resolved spec — do not modify it. Score findings by severity and blast radius. Return machine-readable output only.

Output Format

Return a single JSON block with a findings array.

1. Read .github/skills/clarification-strategies/SKILL.md for Adversarial Stress-Test Patterns and Adversarial Scoring Protocol.
2. Read the spec at SpecPath. Detect spec_type from frontmatter (default: product).
3. Extract all cross-referenceable artifacts:
   • Requirement IDs (FR-###, TR-###, OR-###, RR-###) with their constraint text.
   • Success criteria (SC-###) with their measurable targets.
   • Work items (user stories US# or objectives OBJ#) with acceptance/validation/verification scenarios.
   • Scope boundaries (Included, Excluded, Edge Cases & Boundaries).
   • Constraints (non-functional requirements, technical/operational constraints, assumptions).
4. Run four detection passes:
   • Cross-Requirement Contradiction: Pair-wise comparison of quantified constraints across all requirement and SC entries. Flag pairs whose stated bounds conflict at any scale within scope.
   • Constraint Impossibility: For each SC, verify the combined constraint set (performance, uptime, scope exclusions, deployment model) has a feasible solution given stated requirements.
   • Concurrent-Trigger Ambiguity: Identify work-item pairs sharing an actor or trigger context with no defined ordering, mutual exclusion, or conflict-resolution rule.
   • Boundary/Scale Stress: For every quantified constraint, check for 0, max, and max+1 test scenarios. For every unconstrained resource, flag the absence of a bound.
5. Score each finding per the Adversarial Scoring Protocol:
   • Assign severity: CRITICAL, HIGH, or MEDIUM.
   • Count blast radius (number of distinct affected IDs).
   • Rank by severity × blast_radius (CRITICAL=3, HIGH=2, MEDIUM=1).
   • Cap at 5 findings. Drop lowest-ranked beyond the cap.
6. Allocate IDs only after ranking and capping findings:
   • New findings start at HighestFindingNumber + 1 and increase monotonically.
   • Never reuse a gap or any ID in ExistingFindingIds.
   • Before returning, compare every allocated ID against ExistingFindingIds and all other returned IDs. On any collision, return the collision error below and no findings; never renumber a persisted finding.
7. Return a single JSON block: