What makes adversarial-plan notable?

2.7K downloads. Trust evidence available

How should adversarial-plan 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-plan

Adversarial implementation planner. Takes a spec.md (from adversarial-spec) and optionally review findings, then produces a plan.md with ordered steps, dependencies, files, tests, and risks. Execute the result through focused per-step specs. Skill: adversarial-plan Owner: chpomob Summary: Adversarial implementation planner. Takes a spec.md (from adversarial-spec) and optionally review findings, then produces a plan.md with ordered steps, dependencies, files, tests, and risks. Execute the result through focused per-step specs. Tags: latest:0.1.0 Version history: v0.1.0 | 2026-08-03T18:11:42.037Z | auto Initial release of adversarial-plan: an adversarial i

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-plan

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-plan/snapshot"

Documentation

CLAWHUB

47,644 characters of source documentation, loaded on request.

Extracted files

5 files captured from the source.

SKILL.md

---
name: adversarial-plan
description: "Adversarial implementation planner. Takes a spec.md (from adversarial-spec) and optionally review findings, then produces a plan.md with ordered steps, dependencies, files, tests, and risks. Execute the result through focused per-step specs."
version: 1.0.0
author: Hermes Agent
license: 0BSD
platforms: [linux, macos]
metadata:
  hermes:
    tags: [adversarial, planning, implementation, plan, architecture]
    related_skills: [adversarial-spec, adversarial-code-loop, adversarial-code-review]
---

# Adversarial Plan

**Spec → implementation plan.** Two-role adversarial pipeline that takes a spec.md (from
adversarial-spec) and optionally review findings (from adversarial-code-review), and
produces a plan.md with ordered steps. To implement it with
adversarial-code-loop, convert each step to a focused spec and run the steps in
dependency order; adversarial-code-loop has no functional `--plan` mode.

## Installation

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

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

or, from an existing checkout:

bash scripts/install.sh

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

## Workflow

```text
PREFLIGHT ──→ optional DEEP RESEARCH ──→ GIT SETUP
                                           │
                                           ├─ delegated success ──→ FINALIZE
                                           │
                                           └─ direct/fallback ──→ PLAN ──→ CHALLENGE
                                                                          │
                                               APPROVE + no findings ──────┤
                                                                          │
                                               otherwise ──→ REVISE ──→ VERIFY
                                                                  ↑          │
                                                                  └──────────┘ up to --max-loops
                                                                          │
                                                                      FINALIZE
```

There is one CHALLENGE phase and no `CROSS_2` phase. A successful delegated
run bypasses PLAN/CHALLENGE/REVISE/VERIFY; a delegated fallback enters the
normal adversarial loop. FINALIZE squash-merges an approved plan unless
`--no-merge` is set, or records a rejection marker when findings remain.

## Prompt design

The CHALLENGE prompt references `plan.md` and `spec.md` on disk and instructs
the challenger to read them from the current working directory (the phase
workdir). No document text is embedded in the prompt; the provider runs with
the phase workdir as its cwd, so filesystem-capable providers inspect both
files and the cumulative branch diff directly.

## CLI

<!-- CL

README.md

# adversarial-plan

**Spec → implementation plan.** Two-role adversarial pipeline that reads a spec (from `adversarial-spec`) and optional review findings (from `adversarial-code-review`), then produces a `plan.md` with ordered steps.

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

## How it works

```text
PREFLIGHT ──→ optional DEEP RESEARCH ──→ GIT SETUP
                                           │
                                           ├─ delegated success ──→ FINALIZE
                                           │
                                           └─ direct/fallback ──→ PLAN ──→ CHALLENGE
                                                                          │
                                               APPROVE + no findings ──────┤
                                                                          │
                                               otherwise ──→ REVISE ──→ VERIFY
                                                                  ↑          │
                                                                  └──────────┘ up to --max-loops
                                                                          │
                                                                      FINALIZE
```

The direct pipeline has one CHALLENGE phase and no `CROSS_2` phase. A successful
delegated run bypasses the adversarial loop; delegated fallback uses the direct
pipeline. FINALIZE squash-merges approval unless `--no-merge` is set, or records
a rejection marker if findings remain.

## Plan format

```yaml
### P1: Step title
- **Files:** /path/to/file1, /path/to/file2
- **Description:** What to implement
- **Dependencies:** []
- **Tests:** How to verify
- **Risks:** What could go wrong
```

`adversarial-code-loop` does not provide a functional plan-file mode. To
implement a generated plan, convert each step's Files, Description, Tests, and
Risks into a focused spec, then invoke the code loop with `--spec` for each step
in dependency order. See [Running plan steps without plan mode](references/run-plan-steps-without-plan-mode.md)
for the complete workflow and launch example.

## Comparison

| Feature | adversarial-plan | Manual planning |
|---------|-----------------|-----------------|
| Adversarial challenge | ✅ plan-challenger critiques order, gaps, risks | ❌ |
| Git-native | ✅ branch-per-plan, squash-merge | ❌ |
| Findings-aware | ✅ accepts structured findings JSON | ❌ |
| Code-loop workflow | ✅ per-step specs feed repeated `--spec` runs | Manual step extraction |

## Quick start

```bash
python3 scripts/adversarial_plan.py \
  --spec spec.md \
  --findings findings.json \
  --dev-cmd "pi --provider zai --model glm-5.2" \
  --review-cmd "pi --provider deepseek --model deepseek-v4-pro"
```

## Dependencies

- Python ≥ 3.11
- Git ≥ 2.5
- Two LLM CLIs (plan-writer + plan-challenger)

Uses `adversarial-common` as the shared engine.

## License

0BSD — see [LICENSE](LICENSE).

_meta.json

{
  "ownerId": "kn7e26az9x7m8bgwfwg90q1wkh8bsqw0",
  "slug": "adversarial-plan",
  "version": "0.1.0",
  "publishedAt": 1785780702037
}

references/claude-timeout-notes.md

# Claude Fable 5 Timeout Notes

When using Claude Fable 5 as plan-challenger:

- Extended thinking takes 8-12 min per response
- The default adversarial-plan timeout of 600s is often insufficient
- Increase `--timeout` to at least 1200 when Claude is the `--review-cmd`
- Pair with `--hard-timeout 1800` inside the claude-tmux command
- If Claude exits code 3 (REJECT) due to non-parseable JSON, retry with DeepSeek or GLM-5.2
- Validated 2026-07-10: 4 findings, REQUEST_CHANGES → REVISE → APPROVE, 4/4 settled

references/codex-claude-full-pipeline.md

# Codex GPT-5.6-Sol + Claude Fable 5 — Full Adversarial Pipeline

Validated 2026-07-10 on the OmniSense firmware project (ESP32-S3 + CC1101).

## Pipeline stages

All three stages completed in 1 cycle each with Codex as writer/DEV and Claude as
challenger/reviewer:

| Stage | Writer | Challenger | Findings | Result |
|-------|--------|------------|----------|--------|
| Spec | Codex GPT-5.6-Sol (reasoning=high) | Claude Fable 5 (tmux) | 11 findings | APPROVED (11/11 settled) |
| Plan | Codex GPT-5.6-Sol (reasoning=high) | Claude Fable 5 (tmux) | 4 findings | APPROVED (4/4 settled) |
| Code loop | Codex GPT-5.6-Sol (reasoning=high) | Claude Fable 5 (tmux) | 4+ findings per step | In progress (P5/8 reached) |

## Commands used

### Spec
```bash
python3 adversarial_spec.py \
  --brief /tmp/brief.md \
  --dev-cmd "codex exec -C /path/to/target-repo --skip-git-repo-check --dangerously-bypass-approvals-and-sandbox -c model='gpt-5.6-sol' -c model_reasoning_effort='high'" \
  --review-cmd "python3 /path/to/claude-tmux.py --timeout 600 --hard-timeout 1200 --cwd /path/to/target-repo" \
  --feature "feature-name" --timeout 1200
```

### Plan
```bash
python3 adversarial_plan.py \
  --spec spec.md \
  --dev-cmd "codex exec -C /path/to/target-repo ..." \
  --review-cmd "python3 /path/to/claude-tmux.py --cwd /path/to/target-repo ..." \
  --feature "feature-name" --timeout 1200
```

### Code loop
```bash
python3 adversarial_loop.py \
  --plan /tmp/plan.md \
  --dev-cmd "codex exec -C /path/to/target-repo ..." \
  --review-cmd "python3 /path/to/claude-tmux.py --cwd /path/to/target-repo ..." \
  --feature "feature-name" --out .adversarial-loop \
  --timeout 1200 --max-loops 2 --no-arbiter
```

## Key observations

- Claude Fable 5 via `claude-tmux.py` produced valid JSON for both the embedded-prompt
  pattern (spec/plan challenger) and the files-on-disk pattern (code loop reviewer).
  Earlier documentation claiming Claude cannot do the embedded-prompt pattern was
  pre-Fable-5 or related to an older claude-tmux wrapper version.
- claude-tmux wrapper buffers ALL output until the session completes — no partial
  output appears in `process(action='poll')`. Only `notify_on_complete` reveals the result.
- Codex with `codex exec -C <dir>` + inline prompt (`-C` for context directory,
  prompt as argument) is the preferred approach for focused reviews, avoiding the
  1 MB stdin input limit of the review script's `--project-dir` mode.
- `reasoning=high` on GPT-5.6-Sol produces deeper analysis but can cause 5+ minute
  silent pauses between actions. `reasoning=low` is faster for exploration-heavy tasks.
- The full pipeline produces real git commits at every stage, making rollback safe.
- Claude quota is the main bottleneck: ~200-300K tokens per 5h window. For long
  code loops (8+ steps), GLM-5.2 (pi --provider zai) or DeepSeek are viable
  fallbacks for the reviewer 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-plan",
      "sourceUrl": "https://clawhub.ai/chpomob/skills/adversarial-plan",
      "sourceType": "profile",
      "confidence": "medium",
      "observedAt": "2026-10-09T11:52:46.409Z",
      "isPublic": true
    },
    {
      "factKey": "protocols",
      "category": "compatibility",
      "label": "Protocol compatibility",
      "value": "OpenClaw",
      "href": "https://www.xpersona.co/api/v1/agents/clawhub-chpomob-adversarial-plan/contract",
      "sourceUrl": "https://www.xpersona.co/api/v1/agents/clawhub-chpomob-adversarial-plan/contract",
      "sourceType": "contract",
      "confidence": "medium",
      "observedAt": "2026-10-09T11:52:46.409Z",
      "isPublic": true
    },
    {
      "factKey": "traction",
      "category": "adoption",
      "label": "Adoption signal",
      "value": "2.7K downloads",
      "href": "https://clawhub.ai/chpomob/adversarial-plan",
      "sourceUrl": "https://clawhub.ai/chpomob/adversarial-plan",
      "sourceType": "profile",
      "confidence": "medium",
      "observedAt": "2026-10-09T11:52:46.409Z",
      "isPublic": true
    },
    {
      "factKey": "latest_release",
      "category": "release",
      "label": "Latest release",
      "value": "0.1.0",
      "href": "https://clawhub.ai/chpomob/adversarial-plan",
      "sourceUrl": "https://clawhub.ai/chpomob/adversarial-plan",
      "sourceType": "release",
      "confidence": "medium",
      "observedAt": "2026-08-03T18:11:42.037Z",
      "isPublic": true
    },
    {
      "factKey": "handshake_status",
      "category": "security",
      "label": "Handshake status",
      "value": "UNKNOWN",
      "href": "https://www.xpersona.co/api/v1/agents/clawhub-chpomob-adversarial-plan/trust",
      "sourceUrl": "https://www.xpersona.co/api/v1/agents/clawhub-chpomob-adversarial-plan/trust",
      "sourceType": "trust",
      "confidence": "medium",
      "observedAt": null,
      "isPublic": true
    }
  ],
  "events": [
    {
      "eventType": "release",
      "title": "Release 0.1.0",
      "description": "Initial release of adversarial-plan: an adversarial implementation planner. - Takes a spec.md (from adversarial-spec), optional review findings, and generates a structured plan.md with ordered steps, dependencies, files, tests, and risks. - CLI supports extensive options for workflow control, provider selection, research, and CI integration. - Implements a challenge-revise-verify loop, with support for delegated execution and deep external research. - Outputs plan.md with clear YAML frontmatter and step breakdown; integrates with adversarial-code-loop for stepwise implementation. - Includes HTML reporting, detailed exit codes, and integration guidance. - Personas and related skills organized via adversarial-common for extensibility.",
      "href": "https://clawhub.ai/chpomob/adversarial-plan",
      "sourceUrl": "https://clawhub.ai/chpomob/adversarial-plan",
      "sourceType": "release",
      "confidence": "medium",
      "observedAt": "2026-08-03T18:11:42.037Z",
      "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/claude-timeout-notes.md

references/codex-claude-full-pipeline.md

Related agents

activepieces

cherry-studio

AionUi

CopilotKit

Machine-readable data