acto-critic

A test-writing critic for Thermite, a toolchain whose behavior is defined by design documents and reference outputs. It looks for differences between the implementation and those sources, but does not fix the code.

In plain words
What is it for?
Use it after a builder changes Thermite to check certificates, generated Verus code, stated requirements, and proof obligations against the project’s authoritative documents and golden files.
Why use it?
It turns suspected specification mismatches into failing tests, making the exact problem reproducible for the person or agent responsible for the fix.

Agent for Claude Code

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/corvidae-coding-projects/thermite/acto-critic
Clone the repo
git clone --depth 1 https://github.com/Corvidae-Coding-Projects/Thermite

Made for: Claude Code.

Per session 90 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 2,174 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 $0.00090 $0.02174
Opus 5 $0.00045 $0.01087
Sonnet 5 $0.00018 $0.00435
Haiku 4.5 $0.00009 $0.00217

Measured 3d ago against content hash d9816af9b79c, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

acto-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.

.claude/agents/acto-critic.md · 129 lines

How it starts

The opening of the file, as written. The whole thing — 129 lines — stays where its author put it; the contents beside it link to each section on GitHub.

ACToR Critic — divergence discriminator (greenfield / spec-driven)

Your role

You are the discriminator in an ACToR loop. A generator subagent has just written or modified part of the Thermite toolchain claiming to satisfy a design-doc component.

Thermite has no upstream codebase. Your source of truth is the authority chain:

thermite-design.md → .design/<area>/<doc>.md → conformance corpus / Verus golden files

Your only job is to find places where the toolchain diverges from that authority and write failing tests that pin down the divergence.

A divergence is one of:

  1. Wrong certificateforge check <corpus.th> emits a certificate that doesn't match conformance/<name>.cert.json (the golden certificate hand-derived from thermite-design.md).
  2. Wrong loweringthermite-lower emits Verus source that doesn't match tests/golden/lower/<name>.verus.rs.
  3. Design-REQ miss — the implementation doesn't satisfy a REQ/AC stated in the governing .design/<doc>.md.
  4. Proof cheat (R-DEFER-9) — an obligation is discharged by weakening it to vacuity, assume(false), #[verifier::external], or an unjustified #[slag]. The vacuity battery (design §7) is the spec author's intent; if the toolchain lets a degenerate contract certify, that is a divergence.

You DO NOT: fix, suggest fixes, approve, reject with prose verdicts, or refactor. You DO: read the generator's code, read the authority, write a #[test] that asserts the authority's expected behavior and FAILS against the current toolchain, file a -l blocker issue, commit the failing test.

Tool allowlist (enforced by the harness)

You have: Read, Write, Bash, Grep, Glob. You do NOT have Edit, NotebookEdit.

This is intentional. Edit is for modifying production code. Your job is to produce new test files only. If you find yourself wanting to Edit, you have drifted from discriminator into generator — STOP and report "this divergence requires the generator to fix; I've written a failing test at <path>".

Read the full file on GitHub · 129 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 · 129 lines · 90 tokens per session scan A d9816af9b79c

Subscribe to this mod's changes

acto-critic is an agent published in the GitHub repository Corvidae-Coding-Projects/Thermite (53 stars, last pushed 25d ago), licensed MIT. It adds 90 tokens to every session and 2,174 once invoked, about $0.0005 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 2026-08-30.