| name | adversarial |
| description | Generative-adversarial coordination — locks a generator and critic agent in an escalating quality loop. The critic actively tries to break the generator's output with increasing sophistication. Quality emerges from adversarial pressure, not checklist compliance. Use when you need deep quality hardening beyond standard review, or when the user invokes /adversarial run <spec-path>. |
Generative-Adversarial Skill
Iterative GENERATE ↔ ATTACK loop with escalating critic modes for deep quality hardening.
Usage
/adversarial run <spec-path>
Example:
/adversarial run specs/044-factory-skill-mvp/README.md
Execution
This skill is orchestrated by the pipeline engine. When invoked, execute:
synodic harness run --pipeline adversarial --spec <spec-path>
Pipeline definition: .harness/pipelines/adversarial.yml
Prompt templates: skills/adversarial/prompts/
Output schemas: schemas/generate-report.json, schemas/attack-report.json
Pipeline Steps
- ADVERSARIAL LOOP (fan/loop, max 5 iterations):
- GENERATE — Implement spec + harden against anticipated attacks (worktree)
- STATIC GATE — Preflight checks (cap 2 rework)
- ATTACK — Adversarial critic tries to break output (read-only)
- Termination — 2 consecutive clean rounds OR max rounds OR quality plateau
- CREATE PR — Push and create pull request
Escalation Ladder
Critic modes escalate progressively:
syntax-and-types — Type errors, API misuse, obvious bugs
edge-cases — Boundary conditions, empty inputs, overflow
concurrency-safety — Race conditions, deadlocks, data races
adversarial-inputs — Injection, malformed data, fuzzing
semantic-analysis — Logic correctness, invariant violations
Governance
- Checkpoint map: STATIC GATE → Layer 1, ATTACK → Layer 2, Plateau → escalate
- Governance log:
.harness/adversarial.governance.jsonl
- See HARNESS.md for the governance protocol