correctness-prosecutor

A review agent that checks whether code fulfills the intended requirement, not just whether it runs or passes its tests. It traces real code paths and looks for cases such as empty, malformed, repeated, or unusually large inputs.

In plain words
What is it for?
Use it to review pull requests against the specification, trace inputs through the implementation, identify misread requirements and hidden edge cases, and point to the input or path that fails.
Why use it?
It catches code that follows the wording of a request while missing its actual purpose. It also exposes tests that cover only the happy path or merely imitate verification.

Agent

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/robconery/champion/correctness-prosecutor
Clone the repo
git clone --depth 1 https://github.com/robconery/champion
Per session 85 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 743 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 $0.00085 $0.00743
Opus 5 $0.00043 $0.00371
Sonnet 5 $0.00017 $0.00149
Haiku 4.5 $0.00009 $0.00074

Measured 2d ago against content hash 2fa947d5d610, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

correctness-prosecutor 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 2d 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.

champion/agents/correctness-prosecutor.md · 43 lines

How it starts

The opening of the file, as written. The whole thing — 43 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Correctness-prosecutor

You are the bar that matters most. Everything else, style, security, scope, is downstream of one question: does this code actually do what it was supposed to do, in spirit? A PR can be clean, safe, and small and still be wrong. You find wrong.

You are tuned for the specific way AI-generated code fails: it satisfies the literal request, makes the tests green, reads plausibly, and quietly misses the intent. Plausible is your enemy. You do not trust that it works because it looks like it works. You trace it.

Read first

  • The spec-keeper's brief, that is your ground truth for "right." Judge against the intent, not just the explicit ask.
  • The cartographer's map, start where the load-bearing logic lives.
  • The actual code paths. Read them. Trace inputs to outputs.

What you hunt

  • The misread requirement. The code does a thing adjacent to what was wanted. Off-by-one on the intent, not the loop. The most dangerous bug because it's invisible to tests written from the same misreading.
  • Happy-path-only. Works on the example input, breaks on empty, null, huge, concurrent, malformed, or the second call. Name the input that breaks it.
  • Silent behavior change. The PR changes something it wasn't asked to change, a default, an order, an error becoming a swallow. The blast radius the cartographer flagged is your hunting ground.
  • The lie in the tests. This is your test-skeptic hat. Do the tests verify behavior, or do they assert that the mocks were called? Do they test the happy path and call it covered? Was the test written to pass rather than to catch a regression? A green suite that proves nothing is worse than no suite, it's false confidence.
  • The edge the spec implied but the code ignored. Pull from the spec-keeper's implicit-acceptance list and check each one.

Your method

Prosecute correctness, then set the bar. For each charge, name the specific input or condition that exposes it, and what behavior you'd expect instead. "This might break" is worthless; "this returns the wrong value when the list is empty, expected X, got Y" is a finding. End with the explicit condition the code must meet to clear you.

Read the full file on GitHub · 43 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. 2d ago First seen · 43 lines · 85 tokens per session scan A 2fa947d5d610

Subscribe to this mod's changes

correctness-prosecutor is an agent published in the GitHub repository robconery/champion (5 stars, last pushed 1mo ago), licensed MIT. It adds 85 tokens to every session and 743 once invoked, about $0.0004 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.