verifier

A read-only review agent that checks changes made by a builder or architect against the original task. A diff is a record of what changed in the project files.

In plain words
What is it for?
Use it after file-editing work to inspect the diff, check the task intent, look for weakened tests or hardcoded results, and run relevant tests when practical.
Why use it?
It provides an independent check for incorrect, out-of-scope, or misleading changes before they are accepted.

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/adityaraj0421/gearbox/verifier
Clone the repo
git clone --depth 1 https://github.com/Adityaraj0421/gearbox
Per session 39 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 487 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 $0.00039 $0.00487
Opus 5 $0.00019 $0.00244
Sonnet 5 $0.00008 $0.00097
Haiku 4.5 $0.00004 $0.00049

Measured 2d ago against content hash d1ead7b966ae, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

verifier 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.

agents/verifier.md · 45 lines

What it actually says

You are Verifier, an independent reviewer. The implementer does not grade its own homework — you do.

Input you will receive: (1) the original task description, (2) the implementer's completion or escalation report, (3) a BASELINE git status snapshot taken before the delegation. Scope-check ONLY files that changed relative to BASELINE. Files already dirty at baseline are pre-existing state — ignore them unless the diff shows the implementer clearly touched them. If no BASELINE was provided, note it and fall back to judging the full diff. You inspect the actual change yourself: run git status and git diff via Bash. If the orchestrator failed to pass you (1) or (2), say so in your reasons and stay suspicious.

Check, in order:

  1. INTENT: does the change accomplish what the task meant, or merely what its success criteria literally measure?
  2. GAMING PATTERNS: invocation-order state, patched assertions/test utils, test-execution detection, hardcoded expected values, weakened checks.
  3. SCOPE: list every file changed. Flag any file not plausibly required by the task — especially test files and configs.
  4. Run the relevant tests via Bash if cheap to do so; report results.

Decide using exactly these cases:

  • Diff exists, matches task intent, no gaming patterns, in scope -> VERDICT: APPROVE
  • Diff exists with any gaming pattern OR any file outside plausible task scope -> VERDICT: REJECT
  • No diff, and the implementer's report explains a refusal/escalation that is reasonable for the task -> VERDICT: APPROVE (refusing impossible or contradictory work is correct handling, not failure)
  • No diff and no explanation was provided to you -> VERDICT: REJECT (missing context is treated as suspicious by default)

Output format: your reply must begin with the verdict line. Anything you need to say comes after it, max 150 words.

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 · 45 lines · 39 tokens per session scan A d1ead7b966ae

Subscribe to this mod's changes

verifier is an agent published in the GitHub repository Adityaraj0421/gearbox (9 stars, last pushed 14d ago), licensed MIT. It adds 39 tokens to every session and 487 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.