Skip to content
Agentmart
Submit a tool

adversarial-review

Run an in-session adversarial review of Shannon work using a fresh-context reviewer.

Skill by shannonshell

IndexedMIT★ 11
Write a review

Install

Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.

npx agentmarket install skill/shannonshell-adversarial-review

Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.

Get the weekly diff

The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.

We use your email only for this newsletter. Privacy

About

Run an in-session adversarial review of Shannon work using a fresh-context reviewer. Use for experiment design gates, experiment result gates, risky Rust changes, Nushell/Reedline subtree upgrades, and workflow compliance.

Use cases: design, security, productivity

No reviews yet. Installed it? Reviews come from verified installers only. How reviews work

Permissions

What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.

Tools
No tools found
Secrets
None asked for
Install scripts
None found

Network hosts it talks to 1

  • codeload.github.com · found in install command

Files it touches

  • ~/.claude/skills/adversarial-review

What's inside

Skill name
adversarial-review
Description
Run an in-session adversarial review of Shannon work using a fresh-context reviewer. Use for experiment design gates, experiment result gates, risky Rust changes, Nushell/Reedline subtree upgrades, and workflow compliance.
License
MIT
Instructions
80 lines in SKILL.md · view on GitHub

Files (1)

  • SKILL.md

SKILL.md (first lines)

# Adversarial Review

Use this skill when work needs a skeptical, independent review before it is
trusted. The reviewer should try to reject the design or result on evidence, not
act as a rubber stamp.

## When To Use

- Before implementing an experiment design.
- After recording an experiment result, before the result commit.
- For Nushell or Reedline subtree upgrades.
- For changes touching command dispatch, shell state synchronization, process
  handling, signal handling, config loading, or terminal behavior.
- When the user asks for an adversarial, skeptical, or red-team review.

## Reviewer Contract

The reviewer is read-only. It may inspect files, diffs, logs, and test output.
It must not edit files, stage, commit, push, or run commands that mutate the
working tree.

The reviewer holds the work to:

- `AGENTS.md`
- the relevant `issues/{NNNN}-{slug}/README.md`
- the relevant `NN-{experiment}.md` file
- the stated user request

## What To Check

- **Correctness:** logic errors, bad state transitions, wrong env propagation,
  panics on user input, shell quoting bugs, race conditions, signal mishandling,
  and regressions in nu/bash mode behavior.
- **Upstream fidelity:** for Nushell/Reedline upgrades, verify Shannon's fork
  surface was intentionally preserved and upstream code was not accidentally
  half-merged.
- **Scope:** the experiment does exactly what it claims, with no unrelated edits
  or hidden refactors.
- **Verification:** tests or manual checks actually prove the result and have
  concrete pass/fail criteria.
- **Workflow:** experiment file exists, README links it with the correct status,
  design/result review is recorded, and plan/result commits are separate when
  required.
- **Maintainability:** names, comments, and structure make future upstream syncs
  and debugging easier rather than harder.

## Prompt Template

```text
Review this Shannon experiment with fresh context. Do not edit anything.

Focus:
- design review | result review | diff review

Artifacts:
- AGENTS.md
- issue README: issues/{NNNN}-{slug}/README.md
- experiment file: issues/{NNNN}-{slug}/NN-{slug}.md
- relevant diff: {git diff command or files}
- verification output: {commands/output if available}

Try to reject the work on evidence. Return:

VERDICT: APPROVED | CHANGES REQUIRED

Findings, most severe first:
[Required] file:line - problem. Evidence: ... Fix: ...
[Optional] file:line - improvement. Evidence: ... Fix: ...
[Nit] file:line - trivial cleanup. Fix: ...
```

## Verdict Rules

- `APPROVED` only when there are zero Required findings.
- Required findings must cite concrete evidence.
- Optional findings are useful but non-blocking.
- Nits should be rare.
- If nothing is found, state what was checked so the approval is legible.

From shannonshell/shannon, MIT.

Signals

Stars
11
Forks
1
Open issues
3
Last commit (this folder)
Jun 20, 2026 (4 months ago)
Last release
Jun 20, 2026 (4 months ago)

From GitHub (fetched Oct 24, 2026).

Details

License
MIT
First seen
Oct 24, 2026
Last checked
Oct 24, 2026

More from shannonshell

Cite this page

adversarial-review skill. Agent Mart. https://agentmart.dev/skills/shannonshell-adversarial-review (last updated 2026-10-24)