Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add arcasilesgroup/ai-engineering --skill ai-adversarial-loopgit clone --depth 1 https://github.com/arcasilesgroup/ai-engineeringWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/arcasilesgroup/ai-engineering/ai-adversarial-loop)<a href="https://agentmods.dev/skills/arcasilesgroup/ai-engineering/ai-adversarial-loop"><img src="https://agentmods.dev/badge/skills/arcasilesgroup/ai-engineering/ai-adversarial-loop/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/arcasilesgroup/ai-engineering/ai-adversarial-loop"><img src="https://agentmods.dev/badge/skills/arcasilesgroup/ai-engineering/ai-adversarial-loop.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00108 | $0.01148 |
| Opus 5.5 | $0.00043 | $0.00459 |
| Sonnet 5.5 | $0.00022 | $0.00230 |
| Haiku 4.5 | $0.00011 | $0.00115 |
Grade A, and why
ai-adversarial-loop scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 19d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 51 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Adversarial loop
You are the coordinator. You run in the main session because only the main session can start and resume agents. Subagents can't launch other agents, and agent-team teammates can't be the lead. You never review code or fix it yourself, and you never read the thread file. You pass turns between the two agents and act on their short returns.
Setup
- Thread file: under
/ai-orchestrator, it's.ai-engineering/workflow/reviews/<slug>-<id>.md. Standalone, it's.ai-engineering/workflow/reviews/<topic>-<YYYY-MM-DD>.md. Create it with a header: the checkpoint's goal and acceptance criteria (or the user's scope), thechanged_files, and the verify commands. Nothing else. - Scope: the checkpoint slice and
changed_files. Standalone, it'sgit diff HEADplus untracked files, and the verify commands are the full test suite, lint and build from the Project config section ofAGENTS.md. - Under
/ai-orchestrator, prompts start with the gate markers:[checkpoint <slug>#<id> review]for the critic and[checkpoint <slug>#<id> fix]for the fixer.
Rounds (at most 4)
- Critic, round 1: start
adversarial-reviewerwith the scope, the thread path and "Debate mode, round 1". Keep its agent ID. If it returnsVERDICT: PASS, you're done; go to the end. - Fixer, round N: start
fixerwith the thread path, the scope and "round N". On later rounds, resume the same fixer with SendMessage and its ID: "Round N: the critic has ruled, see the thread." Add itsFILEStochanged_files. - Critic, round N+1: resume the same critic with SendMessage: "Round N+1: the fixer has responded, see the thread. Verify, rule, and attack the new changes."
VERDICT: PASSends the loop. Otherwise repeat from step 2.
Unlike the orchestrator's "fresh workers" rule, each role here keeps its context: the critic remembers what it flagged, and the fixer remembers what it changed. They stay isolated from each other and from you, and the thread file is the only thing they share.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 19d ago First seen · 51 lines · 108 tokens per session scan A bda5c3a60bff
ai-adversarial-loop is a skill published in the GitHub repository arcasilesgroup/ai-engineering (61 stars, last pushed 2d ago), licensed Apache-2.0. It adds 108 tokens to every session and 1,148 once invoked, about $0.0004 per session on Opus 5.5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-09-27.
Other skills, from other repositories
pipeline-validator
Discovers, triggers, and monitors Azure DevOps pipelines (PR, Buddy Build, Buddy Release) for the current repo and branch. Auto-diagnoses failures from build logs, applies fixes, commits, pushes, and re-triggers until all pipelines pass or max retries reached. Validates PR existence and description completeness.…
ai-agents-empirical-probe-toolkit
Prove-it methods for this repo. Six recipes for runtime-contract probes, guard and threshold calibration, eval A/B, docs-vs-reality audits, reproduce-on-main CI triage, and negative-control test design, each with a worked example from repo history. Use when you say probe the runtime contract, calibrate this guard…
taste-lints
Use when you say run taste lints, check file size, or lint taste invariants. Custom lints with agent-readable remediation instructions for file size, naming conventions, structured logging, and complexity.
ai-agents-debugging-playbook
Symptom-to-triage playbook for this repo's recurring failures. Blocked pushes, drift gate reds, plugin bump reds, coverage pin trips, hook exit 143, session NONCOMPLIANT. Maps each symptom to a first command, discriminating experiment, fix path, and trap. Use when you say triage this failure, why is my push blocked…
ai-agents-config-catalog
Catalog of every configuration axis in this repo, env vars, commit markers, frontmatter keys, QA skip verdicts, and escape hatches, each with its enforcement point and abuse story, plus the checklist for adding a new flag safely. Use when you say what does a skip flag do, list escape hatches, can I skip this gate, add…
extension-audit-methodology
Systematic 5-dimension audit framework for VS Code extensions — debug hygiene, dead code, performance, menu validation, dependency cleanup.