AGENT SKILL

adversarial-breaker

Adversarial verification — fixer→breaker chain. Independent offensive agent.

License 许可
Apache-2.0
最近更新
2026-11-10
Stars
★ 0
Forks
0

01 前言 Front Matter

name
adversarial-breaker
triggers
adversarial, breaker, !breaker, verify fix, romper, verificar fix, try to break, offensive verification
changelog
docs/ciclos/cycle28-20260815.md
description
Adversarial verification — fixer→breaker chain. Independent offensive agent.
token_budget
3000

02 资源文件 5

03 SKILL.md 正文

When to Use

fixer done+ROJA/AMARILLA(auth/storage/API)+not config-only. Skip:VERDE·non-code·comments·docs. Token-save:diff<10L+config/lock/docs.

Protocol

1.Bundle:diff+files+fixer_claims+test_results+zone+pipeline_mode 2.Engram:mem_search("breaker/{target}")→## Past Attack Context 3.Zone:review-rules.jsonc. AMARILLA+auth→ROJA 4.Profiles: .agents/attack-surface.{project}.md→inject 5.Pre-reg:git stash→tests→pop. Break→PRE-BREAK 6.Dispatch:diff>50/ROJA→Parallel Break-7(3). Else→single 7.Launch:breaker-briefing+profile(s)+past+bundle 8.Parse:4-field. Parallel→merge. <3→FAIL→rerun 9.Calibrate:depth/relevance/coverage/specificity 10.Verdict:adjusted+PRE-BREAK 11.R2:FIX→round 2 12.Record:Engram+test cases

Verdicts

AllPASS→APPROVED|1-2minor→FIX|Critical→BLOCK|<3/malformed→FAIL|Timeout→ESCALATE AB-{t}|R:{N}/2|A:{n}|SAFE/BROKEN|V:{v}

R2&Esc

new diff after fix; both diffs—R1+new. ##R2—Focus:{R1}. Max2. Escalate:Target/Rounds/Pipeline/Chain/Blocker/Rec

Integration

quality-gate(BEFORE)|triple-verify(AFTER)|judgment-day|subagent-isolation|external-auditor(BLOCK→confirm)|immune-system(repeated→fix) Projects:.agents/attack-surface.{project}.md+.agents/breaker-profiles/{project}/

Anti-Patterns

Breaker trusts fixer·No attempts·3+r·Before QG·4R review·Happy path·Non-code/tiny diffs·No pre-reg·No test cases·Generic surface

docs/skills/adversarial-breaker/reference.md

Anti-Rationalization

Rationalization Red Flag Verification
"el breaker ya pasó el profile scan, no hace falta re-verificar" Reuse de profile sin fresh scan por cambio Re-ejecutar attack-surface.{project}.md injection + mem_search breaker/{target} por cada diff
"breaker y fixer pueden ser el mismo agente, es más eficiente" Mismo agente verifica su propio fix Delegar breaker independiente vía subagent-isolation con bundle separado, nunca reutilizar fixer context
"clearance permanente tras un PASS" Asumir aprobación vale para cambios futuros Clearance es por-cambio: nuevo diff → nuevo ciclo breaker con verdict fresco AB-{t}

Red Flags

  • Doing work without checking output format → STOP, re-read skill
  • Second occurrence of same rationalization → force RED zone

Verification

  • Output matches skill ## Output contract + file:line citaton
  • cross-ref-check.ps1 → SKILL.md OK

Refs

Cross-Refs: security-scanner | best-practices | judgment-day

04 同仓库更多 Skills