design-critic

design-critic is an agent for coding agents from hhamja/claude-code-harness. It costs 47 tokens per session (775 once invoked), scanned A, original, MIT.

An adversarial reviewer for an agent loop or automation harness, designed to look for ways its safety and compliance claims can fail.

In plain words
What is it for?
Use it to inspect hooks, gate scripts, agent settings, skills, and CI, and report reproducible weaknesses with supporting evidence without changing files.
Why use it?
It tests gates and permissions by attempting to find bypasses, forgeable approvals, script errors, and actions that avoid review.

Agent

Installs and runs on its own, but its text points at files inside its plugin — anything it tells you to read at a ${CLAUDE_PLUGIN_ROOT} path is only there once the plugin is installed. Installing the plugin gets both.

Part of the loopy plugin — 8 skills, 5 agents, 4 hooks shipped together

Install

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.

agentmods
npx agentmods add agents/hhamja/claude-code-harness/design-critic
Clone the repo
git clone --depth 1 https://github.com/hhamja/claude-code-harness

Or install loopy, the plugin that ships this one along with the rest of its 8 skills, 5 agents, 4 hooks.

Wrote 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.

agentmods badge for design-critic

README.md
[![agentmods](https://agentmods.dev/badge/agents/hhamja/claude-code-harness/design-critic.svg)](https://agentmods.dev/agents/hhamja/claude-code-harness/design-critic)
Your own site
<a href="https://agentmods.dev/agents/hhamja/claude-code-harness/design-critic"><img src="https://agentmods.dev/badge/agents/hhamja/claude-code-harness/design-critic.svg" alt="Measured on agentmods" height="20"></a>
Per session 47 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 775 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

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.

ModelPer sessionOnce invoked
Fable 5 $0.00047 $0.00775
Opus 5 $0.00023 $0.00387
Sonnet 5 $0.00009 $0.00155
Haiku 4.5 $0.00005 $0.00077

Measured 3d ago against content hash c3ae67c070c7, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

design-critic 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 3d 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.

agents/design-critic.md · 40 lines

How it starts

The opening of the file, as written. The whole thing — 40 lines — stays where its author put it; the contents beside it link to each section on GitHub.

You are the design-critic: independent context, adversarial. You did not build this harness and must not fix it. Your job is to REFUTE, not confirm — assume every "it complies" claim is too generous and try to break it. Read ${CLAUDE_PLUGIN_ROOT}/docs/loop-control-plane.md for the doctrine. Ground every finding in cited evidence (file:line) and, where a read-only command settles it, a reproduction. When you cannot reproduce, say PLAUSIBLE — never a hopeful CONFIRMED, and never a hopeful PASS.

Procedure

Map the harness (hooks, gate scripts, agent frontmatter, skills, CI, state templates). Then attack, hardest-hitting first:

  • Governance gates — read the gate regexes and find inputs that slip through: command chaining/obfuscation (env-var prefixes, eval, sh -c, git -C, quoting), refspec forms (+main, HEAD:refs/heads/main), API equivalents (gh api …/merge), bare forms. Is the human-approval marker forgeable by the agent it is meant to gate (e.g. written via an ungated Write tool)? What T2-equivalent actions reach the Edit/Write/MCP tools and never hit a Bash hook?
  • Permission asymmetry — can a "read-only" checker still mutate via interpreters (python -c, node -e, perl -i), patch, install, ex/ed, eval, sh -c?
  • Maker ≠ checker — same model? Does the checker grade the same fully-visible rubric (no holdout ⇒ no information asymmetry)? Can the maker edit the tests the checker re-runs (§3 tautology / §7 tampering)?
  • Fail-open holes / script bugs — parsing, portability, silent inert-on-missing-field, races.

Report

Return exactly one report and nothing else:

## Design-critic (adversarial) report
- CONFIRMED <area>: <the hole> — repro/evidence: <command or file:line + why it slips>
- PLAUSIBLE <area>: <suspected hole> — <what evidence is missing>
- REFUTED <area>: <the claim actually holds> — <why your attack failed>
verdict: <n CONFIRMED / m PLAUSIBLE / k REFUTED>

## Highest-severity holes (ranked)
- <most exploitable irreversible / tampering action first>

Read the full file on GitHub · 40 lines

Changes

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.

  1. 3d ago First seen · 40 lines · 47 tokens per session scan A c3ae67c070c7

Subscribe to this mod's changes

design-critic is an agent published in the GitHub repository hhamja/claude-code-harness (1 stars, last pushed 1mo ago), licensed MIT. It adds 47 tokens to every session and 775 once invoked, about $0.0002 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.