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 SSBun/csl-agent-kit --skill adversarial-reviewgit clone --depth 1 https://github.com/SSBun/csl-agent-kitWrote 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/ssbun/csl-agent-kit/adversarial-review)<a href="https://agentmods.dev/skills/ssbun/csl-agent-kit/adversarial-review"><img src="https://agentmods.dev/badge/skills/ssbun/csl-agent-kit/adversarial-review.svg" alt="Measured on agentmods" 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 | $0.00079 | $0.00945 |
| Opus 5 | $0.00039 | $0.00473 |
| Sonnet 5 | $0.00016 | $0.00189 |
| Haiku 4.5 | $0.00008 | $0.00094 |
Grade A, and why
adversarial-review 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 4d 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 — 61 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Adversarial Review
Gate Contract
- Enter only from an explicit user request; never infer the need from risk, complexity, verification gaps, or another workflow.
- Start
BLOCKEDwith distinct Reviewer and Editor agents. - Prefer real isolated subagents for the Reviewer and Editor when the host can dispatch them; fall back to inline role-play with
ISOLATION: simulateddisclosed. Follow Subagent Dispatch. - While
BLOCKED, do not finalize, approve, commit, publish, or externally share the artifact. APPROVEDis review evidence, not external-action authorization.- Set no total round or cycle limit. Continue by review state under Review Loop Contract.
- Keep the active finding ledger in Reviewer–Editor handoffs; do not write report files during intermediate rounds.
- Reviewed artifact changes invalidate
APPROVED; return toBLOCKEDand resume the same numbered review history. - No independent Reviewer: fail closed.
Roles
- Coordinator: Pins scope, gate, routing, and verdict; may be Editor, never Reviewer.
- Reviewer: Read-only; separates findings, required outcomes, and optional remedies under the shared principles.
- Editor: Audits adequacy, minimal resolution, blast radius, and proportionality before accepting, narrowing, rejecting, acknowledging, or escalating; never self-approves.
Reuse the Reviewer. A replacement receives the complete state, artifact, available diff, round number, and finding IDs.
For multiple viable decisions, fixes, or plans, remain BLOCKED and use Decision Consensus Gate before choosing one.
Workflow
0. Resolve dispatch mode
Detect the host's dispatch capability, verify the dispatch path is actually ready (on Pi: the pi-agent carrier is registered and spawnable; on Codex: the host can spawn an agent process), record the mode and per-role readiness once per run, and print the dispatch metadata table to the user as the first output of the run — before any role pass — as specified in Subagent Dispatch. If all roles are ready, enter SUBAGENT mode without asking; if any role is missing, present the table and ask the user whether to proceed in INLINE-FALLBACK, and enter it only on explicit confirmation. Never enter the loop on an unverified dispatch path. The Review Loop Contract, finding ledger, and report format are identical in both modes; APPROVED reached under simulated isolation carries that caveat.
What ships with it
10 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
- evals/report_contract_cases.json 5.3 KB
- evals/semantic_config.json 2.4 KB
- evals/trigger_cases.json 2.4 KB
- examples/agents/adversarial-editor.md 2.9 KB
- examples/agents/adversarial-reviewer.md 2.6 KB
- references/decision-consensus.md 2.1 KB
- references/final-review-report.md 3.1 KB
- references/review-lenses.md 2.4 KB
- references/review-loop.md 6.5 KB
- references/subagent-dispatch.md 14 KB
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.
- 4d ago Changed 2ae4d8c5dc5d
- 8d ago First seen · 61 lines · 79 tokens per session scan A 20d01bd0d635
adversarial-review is a skill published in the GitHub repository SSBun/csl-agent-kit (10 stars, last pushed 7d ago), licensed MIT. It adds 79 tokens to every session and 945 once invoked, about $0.0004 per session on Opus 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 2026-08-31.
Other skills, from other repositories
autoreview
Pre-commit/ship code review: Codex default; optional Claude or Pi.
rework-rate
Measure and interpret PR rework rate — the emerging 5th DORA metric.
omh-code-review
This is a Hermes-native code-review workflow skill.
revdiff-plan
Review the last Codex assistant message (plan, analysis, or proposal) with inline annotations in a TUI overlay. Extracts the most recent response from Codex rollout files and opens it in revdiff for review and annotation. Activates on "revdiff-plan", "review plan with revdiff", "annotate plan", "review last response"…
code-reviewer
Code review specialist focused on patterns, bugs, security, and performance.
agent-teams-simplify-and-harden
Implementation + audit loop using parallel agent teams with structured simplify, harden, and document passes. Spawns implementation agents to do the work, then audit agents to find complexity, security gaps, and spec deviations, then loops until code compiles cleanly, all tests pass, and auditors find zero issues or…