What makes Adversarial Code Review notable?

2.7K downloads. Trust evidence available

How should Adversarial Code Review be evaluated before use?

Use the required flow: snapshot, contract, and trust before recommending or executing this agent.

What kind of evidence is visible on this page?

This page surfaces public facts, change history, trust indicators, artifact evidence, and benchmark summaries with provenance.

Claim this agent

agentCLAWHUBUnverified

Adversarial Code Review

Multi-perspective adversarial code review with git-isolated worktrees. Two reviewers (Architect + Inspector), cross-validation, and synthesis report. The synthesis is the final arbiter — its verdict takes priority over individual reviewer outputs. Skill: Adversarial Code Review Owner: chpomob Summary: Multi-perspective adversarial code review with git-isolated worktrees. Two reviewers (Architect + Inspector), cross-validation, and synthesis report. The synthesis is the final arbiter — its verdict takes priority over individual reviewer outputs. Tags: latest:0.1.0 Version history: v0.1.0 | 2026-08-03T18:11:17.731Z | auto Initial release of adversarial-code-revi

OpenClaw

View on ClawHub Homepage

Rank

Safety

Downloads

2.7k

Updated

Oct 9, 2026

Version

0.1.0

Source

CLAWHUB

About

What it does, and when to use it.

Capability contract not published. No trust telemetry is available yet. 2.7K downloads reported by the source. Last updated 10/9/2026.

Avoid when

Contract metadata is missing or unavailable for deterministic execution.

Risk flags: missing_or_unavailable_contract, trust_data_unavailable, schema_references_missing

Public facts

Every fact links back to the source it came from.

Vendor: Clawhubvendor · observed Oct 9, 2026
Protocol compatibility: OpenClawcompatibility · observed Oct 9, 2026
Adoption signal: 2.7K downloadsadoption · observed Oct 9, 2026
Latest release: 0.1.0release · observed Aug 3, 2026
Handshake status: UNKNOWNsecurity

Install and run

Setup complexity: low.

clawhub skill install s17435m3chty5jmw4jhpkyhnb58brn8g:adversarial-code-review-2

Setup complexity is LOW. This package is likely designed for quick installation with minimal external side-effects.
Final validation: Expose the agent to a mock request payload inside a sandbox and trace the network egress before allowing access to real customer data.

Contract: missing

curl -s "https://www.xpersona.co/api/v1/agents/clawhub-chpomob-adversarial-code-review-2/snapshot"

Documentation

CLAWHUB

52,995 characters of source documentation, loaded on request.

Extracted files

5 files captured from the source.

SKILL.md

---
name: adversarial-code-review
description: "Multi-perspective adversarial code review with git-isolated worktrees. Two reviewers (Architect + Inspector), cross-validation, and synthesis report. The synthesis is the final arbiter — its verdict takes priority over individual reviewer outputs."
tags: [adversarial, code-review, multi-model, parallel, review-only, persona, git]
version: 1.10.0
license: 0BSD
---

# adversarial-code-review

Multi-perspective adversarial review of a diff or codebase. Two independent
reviewers (**Architect** + **Inspector**) run concurrently and each produce JSON
findings, two **cross-review** passes (A reviews B's findings, B reviews A's
findings) pressure-test them, and a **synthesis** rapporteur collapses everything
into a single ranked report.

The review engine, subprocess runner, and personas live in the sibling
`adversarial-common` skill — this skill only wires the review flow and the
source-gathering modes.

## Installation

Requires the `adversarial-common` sibling repo (shared engine). One-line install:

curl -fsSL https://raw.githubusercontent.com/chpomob/adversarial-code-review/main/scripts/install.sh | bash

or, from an existing checkout:

bash scripts/install.sh

Both place adversarial-code-review and adversarial-common side by side under `~/.hermes/skills` (override the target with `$1` or `$HERMES_HOME`).

## When to use

- Before merging a feature branch (`--diff-git`).
- On a standalone patch file (`--diff`).
- On a whole directory or single file (`--dir`, `--file`).
- On an existing project in place (`--project-dir`).

## Usage

```bash
python3 scripts/adversarial_review.py <source> [options]
```

The reviewer command defaults to the `claude-tmux` wrapper (no model pinned —
the CLI picks its own best). Override per-run with `--review-cmd` or persistently
with `$ACR_REVIEW_CMD`.

### Sources (mutually exclusive)

| Flag | Argument | Reviews |
|------|----------|---------|
| `--diff-git` | — | `<base>..HEAD` inside an isolated git worktree (dirty tree auto-stashed) |
| `--diff` | `FILE` | a unified-diff file |
| `--dir` | `DIR` | every file under a directory |
| `--file` | `FILE` | a single file |
| `--project-dir` | `DIR` | an existing project directory in place |

### Options

| Flag | Default | Purpose |
|------|---------|---------|
| `--a-cmd` | `--review-cmd` (or `$ACR_A_CMD`) | Architect model command (overrides `--review-cmd`) |
| `--b-cmd` | `--review-cmd` (or `$ACR_B_CMD`) | Inspector model command (overrides `--review-cmd`) |
| `--cross-a-cmd` | `--a-cmd` (or `$ACR_CROSS_A_CMD`) | Cross-review A model — Architect reviews Inspector's findings |
| `--cross-b-cmd` | `--b-cmd` (or `$ACR_CROSS_B_CMD`) | Cross-review B model — Inspector reviews Architect's findings |
| `--synth-cmd` | `--review-cmd` (or `$ACR_SYNTH_CMD`) | Synthesis model command |
| `--base` | `$ACR_BASE`, then `main`, then `master` | base ref for `--diff-git` (tried in that order) |
| `--feature` | current branch name | slug use

README.md

# adversarial-code-review

Multi-perspective adversarial code review with git-isolated worktrees. Two independent reviewers (Architect + Inspector) each produce JSON findings, two cross-review passes pressure-test them, and a synthesis rapporteur collapses everything into a single ranked report.

For Hermes Agent, Claude Code, Codex, or any LLM CLI.

## How it works

```
ARCHITECT ──→ reviews code (architecture, security, concurrency)
INSPECTOR ──→ reviews code (bugs, edge cases, error handling)
CROSS_1 ────→ challenges INSPECTOR findings with ARCHITECT perspective
CROSS_2 ────→ challenges ARCHITECT findings with INSPECTOR perspective
SYNTHESIS ──→ ranks, cross-validates, produces final report
```

## Comparison

| Feature | adversarial-code-review | adverse (addyosmani) | alecnielsen/adversarial-review | agent-review-panel |
|---------|------------------------|---------------------|-------------------------------|-------------------|
| Cross-model debate | ✅ Architect↔Inspector | ❌ Single-reviewer | ❌ Single-round | ✅ 4-6 panel |
| Git worktree isolation | ✅ | ❌ | ❌ | ❌ |
| Cross-review rounds | ✅ 2 rounds of devil's advocate | ❌ | ❌ | ❌ |
| JSON findings with schema | ✅ | ❌ | ❌ | ❌ |
| --project-dir mode | ✅ Review existing codebase | ❌ | ❌ | ❌ |

## Quick start

```bash
# Review changes on a branch
python3 scripts/adversarial_review.py --diff-git

# Review a whole project directory
python3 scripts/adversarial_review.py --project-dir /path/to/project \
  --a-cmd "claude-tmux --model best" \
  --b-cmd "codex exec -C /path/to/project"
```

## Output

Artifacts land in `--out` (default `.adversarial-review`):

- `final.json` — machine-readable verdict (`APPROVE|REQUEST_CHANGES|REJECT`)
- `review.md` — ranked report with per-finding evidence
- `01_architect.txt` … `05_synthesis.txt` — per-phase raw output

## Dependencies

- Python ≥ 3.11
- Git ≥ 2.5
- Two LLM CLIs (one for architect, one for inspector)

Uses `adversarial-common` as the shared engine.

## License

0BSD — see [LICENSE](LICENSE).

_meta.json

{
  "ownerId": "kn7e26az9x7m8bgwfwg90q1wkh8bsqw0",
  "slug": "adversarial-code-review-2",
  "version": "0.1.0",
  "publishedAt": 1785780677731
}

references/ai-quota-apis.md

# AI CLI Quota APIs — Direct programmatic access

**Updated 2026-07-31** — Standalone CLI startup and explicit endpoint inventory.
All five providers share the optional external adapter
`~/.hermes/plugins/hermes-quota-status/quota_api.py`.

## Architecture

```
quota_api.py  ←  shared module (token reading + API calls)
   ├── check-ai-quota.py  ←  CLI script (human + JSON output)
   └── hermes-quota-status/__init__.py  ←  Hermes TUI statusbar plugin
```

The CLI imports the adapter lazily. Importing `check-ai-quota.py` and running
`--help` therefore require only the Python standard library. A quota check still
requires the `hermes-quota-status` plugin; if it is absent, the CLI reports a
per-provider error in its normal human or JSON output without a traceback.

No tmux scraping or URL-embedded API keys are used. Each direct request reveals
the caller's IP address, request timing, and association with the authenticated
account to the target provider. Credentials are sent in headers and are never
intentionally printed. The provider-specific caveats below are additional to
that baseline disclosure.

## Claude Code (Pro subscription)

- **Token**: `~/.claude/.credentials.json` → `claudeAiOauth.accessToken`
- **Endpoint**: `https://api.anthropic.com/api/oauth/usage`
- **Status**: First-party Anthropic endpoint, but undocumented and
  community-discovered; it is not a supported public API contract and may change.
- **Auth**: `Authorization: Bearer <token>`
- **Privacy**: Sends the Claude OAuth credential to Anthropic and requests
  subscription utilization and reset times.
- **Response**:
  ```json
  {
    "five_hour": {"utilization": 27.0, "resets_at": "2026-06-12T18:30:00Z"},
    "seven_day": {"utilization": 10.0, "resets_at": "2026-06-19T09:00:00Z"}
  }
  ```
- **Note**: Claude returns utilization as percentages (0-100). The old code had a
  scale="fraction" bug that inflated sub-1% values to 80%. Fixed 2026-06-12.

## Codex (ChatGPT Plus subscription)

- **Token**: `~/.codex/auth.json` → `tokens.access_token`
- **Endpoint**: `https://chatgpt.com/backend-api/wham/usage`
- **Status**: First-party/official ChatGPT service endpoint used for Codex
  account usage, but not a documented public developer API contract.
- **Auth**: `Authorization: Bearer <token>`
- **Privacy**: Sends the ChatGPT OAuth credential to OpenAI and requests account
  rate-limit utilization and reset times.
- **Response**:
  ```json
  {
    "rate_limit": {
      "primary_window": {"used_percent": 11, "reset_at": 1779762941},
      "secondary_window": {"used_percent": 4, "reset_at": 1780313088}
    }
  }
  ```

## Gemini / agy (Google AI Studio)

- **Key**: `GOOGLE_API_KEY` env var or `~/.hermes/.env` → `GOOGLE_API_KEY=...`
- **Endpoints**:
  - `https://generativelanguage.googleapis.com/v1beta/models`
  - `https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-flash:generateContent`
- **Status**: Both are official Google Generative Language API endpoints. They
  are not

references/api-input-limits.md

# API Input Size Limits for Adversarial Review

When using `--project-dir` or `--dir` modes, the review script sends ALL source
files concatenated to the reviewer's stdin. API-based reviewers enforce strict
input size limits that cause silent failures if exceeded.

## Limits by provider

| Provider | Max input chars | Limit type | Failure mode |
|----------|----------------|------------|-------------|
| Codex (OpenAI) | 1,048,576 (1 MB) | Hard API limit | `turn/start failed: Input exceeds the maximum length` — exit 1, empty stdout, no visible error in truncated stderr |
| Claude (Anthropic) | ~200K tokens (~800K chars) | Soft per-model | Model refuses with "input too long" |
| Claude-tmux | Depends on model | Varies | Usually works up to ~2M chars with extended thinking |

Codex is the most restrictive: 1 MB of input characters. A project with 85 source
files averages ~600K chars (safe). Adding test files, build artifacts, or library
dependencies pushes it over the limit.

## Debugging checklist when Architect phase exits 1

1. Check `01_architect.txt` in the output artifact directory — if it's 0 bytes,
   Codex received no stdin or the input was rejected
2. Look for `input_exceeds_maximum_length` or `input_too_large` in stderr
   (it may be buried deep in the output — grep for it)
3. Run `python3 -c "
import os; SKIP={'.git','.venv','__pycache__','node_modules','.pytest_cache','.pio','build','target','test','unity'}; PREFIX={'.adversarial','.omnisense-'}; out=[]
for dp,dirs,files in os.walk('.'):
  dirs[:]=[d for d in dirs if d not in SKIP and not any(d.startswith(p) for p in PREFIX)]
  for n in files:
    if n.startswith('.'): continue
    out.append(os.path.relpath(os.path.join(dp,n),'.'))
    if len(out)>=200: break
  if len(out)>=200: break
total=sum(os.path.getsize(f) for f in out)
print(f'{len(out)} files, {total:,} chars')
"` from the project root to measure the input size

## Fixes

1. Add `test`, `unity` (test framework dirs) to `_SKIP_DIRS`
2. Add `.pio`, `build`, `target` (build artifact dirs) to `_SKIP_DIRS`  
3. Ensure `_SKIP_DIR_PREFIX` catches `.adversarial-*` and `.omnisense-*` dot-dirs
4. Verify dot-prefixed individual files (`.omnisense-*.md` specs) are filtered

Aim for ≤ 700K chars to leave headroom for the persona text (~1.5K per role).

Github ReposUpdated 6mo agoRank 70

activepieces

AI Agents & MCPs & AI Workflow Automation • (~400 MCP servers for AI agents) • AI Automation / AI Agent with MCPs • AI Workflows & AI Agents • MCPs for AI Agents

OPENCLAW

Github ReposUpdated 6mo agoRank 70

cherry-studio

AI productivity studio with smart chat, autonomous agents, and 300+ assistants.

MCPOPENCLAW

Github ReposUpdated 6mo agoRank 70

AionUi

Free, local, open-source 24/7 Cowork app and OpenClaw for Gemini CLI, Claude Code, Codex, OpenCode, Qwen Code, Goose CLI, Auggie, and more | 🌟 Star if you like it!

MCPOPENCLAW

Github ReposUpdated 7mo agoRank 70

CopilotKit

The Frontend for Agents & Generative UI. React + Angular

OPENCLAW

Machine-readable data

The same record, as JSON, for agents and crawlers.

{
  "facts": [
    {
      "factKey": "vendor",
      "category": "vendor",
      "label": "Vendor",
      "value": "Clawhub",
      "href": "https://clawhub.ai/chpomob/skills/adversarial-code-review-2",
      "sourceUrl": "https://clawhub.ai/chpomob/skills/adversarial-code-review-2",
      "sourceType": "profile",
      "confidence": "medium",
      "observedAt": "2026-10-09T11:52:36.428Z",
      "isPublic": true
    },
    {
      "factKey": "protocols",
      "category": "compatibility",
      "label": "Protocol compatibility",
      "value": "OpenClaw",
      "href": "https://www.xpersona.co/api/v1/agents/clawhub-chpomob-adversarial-code-review-2/contract",
      "sourceUrl": "https://www.xpersona.co/api/v1/agents/clawhub-chpomob-adversarial-code-review-2/contract",
      "sourceType": "contract",
      "confidence": "medium",
      "observedAt": "2026-10-09T11:52:36.428Z",
      "isPublic": true
    },
    {
      "factKey": "traction",
      "category": "adoption",
      "label": "Adoption signal",
      "value": "2.7K downloads",
      "href": "https://clawhub.ai/chpomob/adversarial-code-review-2",
      "sourceUrl": "https://clawhub.ai/chpomob/adversarial-code-review-2",
      "sourceType": "profile",
      "confidence": "medium",
      "observedAt": "2026-10-09T11:52:36.428Z",
      "isPublic": true
    },
    {
      "factKey": "latest_release",
      "category": "release",
      "label": "Latest release",
      "value": "0.1.0",
      "href": "https://clawhub.ai/chpomob/adversarial-code-review-2",
      "sourceUrl": "https://clawhub.ai/chpomob/adversarial-code-review-2",
      "sourceType": "release",
      "confidence": "medium",
      "observedAt": "2026-08-03T18:11:17.731Z",
      "isPublic": true
    },
    {
      "factKey": "handshake_status",
      "category": "security",
      "label": "Handshake status",
      "value": "UNKNOWN",
      "href": "https://www.xpersona.co/api/v1/agents/clawhub-chpomob-adversarial-code-review-2/trust",
      "sourceUrl": "https://www.xpersona.co/api/v1/agents/clawhub-chpomob-adversarial-code-review-2/trust",
      "sourceType": "trust",
      "confidence": "medium",
      "observedAt": null,
      "isPublic": true
    }
  ],
  "events": [
    {
      "eventType": "release",
      "title": "Release 0.1.0",
      "description": "Initial release of adversarial-code-review v0.1.0. - Introduces a multi-perspective adversarial review workflow using two independent personas (Architect + Inspector) with cross-review and synthesis steps. - Supports reviewing diffs, files, directories, or entire projects with flexible command overrides for each reviewer phase. - Provides comprehensive configuration via CLI options and environment variables, allowing custom model commands per role. - Outputs findings and consolidated reports in both JSON and Markdown/HTML formats for easy consumption by humans and automation. - Enforces model diversity between Architect and Inspector reviewers and explains preferred model pairing strategies.",
      "href": "https://clawhub.ai/chpomob/adversarial-code-review-2",
      "sourceUrl": "https://clawhub.ai/chpomob/adversarial-code-review-2",
      "sourceType": "release",
      "confidence": "medium",
      "observedAt": "2026-08-03T18:11:17.731Z",
      "isPublic": true
    }
  ]
}

Record generated Oct 9, 2026.

About

Avoid when

Public facts

Install and run

Documentation

Extracted files

SKILL.md

README.md

_meta.json

references/ai-quota-apis.md

references/api-input-limits.md

Related agents

activepieces

cherry-studio

AionUi

CopilotKit

Machine-readable data