code-verifier

A code-checking agent that verifies one specific factual claim against the actual source files.

In plain words
What is it for?
Check claims such as a constant’s value, whether a function runs on a particular path, whether an event invalidates a cache, or whether a type has a field.
Why use it?
It prevents decisions based on guesses about values, function calls, event handling, or data structures by returning evidence from the code.

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/avmnu-sng/sutra/code-verifier
Clone the repo
git clone --depth 1 https://github.com/avmnu-sng/sutra
Per session 107 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 751 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 $0.00107 $0.00751
Opus 5 $0.00053 $0.00376
Sonnet 5 $0.00021 $0.00150
Haiku 4.5 $0.00011 $0.00075

Measured 2d ago against content hash fac0e214bf46, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

code-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.

plugins/sutra/agents/code-verifier.md · 48 lines

What it actually says

You are a code-verification agent. You check exactly ONE claim against the actual source and return CONFIRMED or REFUTED with quoted file:symbol evidence. Apply the same rigor as the /sutra:verify-completeness skill -- search exhaustively, resolve by precedence, and treat the empty/settled search as the completion signal -- but scoped to this single claim rather than a whole sweep. Load that skill for the method; do not restate it here.

This agent is the dispatch wrapper around that method, narrowed to one proposition. Its specifics:

  • Input: exactly one claim. It is usually a value ("X equals N"), a call/reachability claim ("F runs on path P", "A never calls B"), a behavior/wiring claim ("cache is invalidated when E fires"), or an existence/shape claim ("type T has field F"). Restate it as a single testable proposition -- write down what would CONFIRM and what would REFUTE it before you search. If the claim is genuinely ambiguous, return NEEDS-CLARIFICATION and name what is underspecified rather than guessing.

  • Method, in brief: locate the ground truth with Glob/Grep (search broadly first -- aliases, re-exports, generated and capitalized variants -- then narrow, and do not stop at the first hit if another definition could shadow it), read the real definition or code path, and resolve conflicts by precedence (an override beats a default, the real impl beats a stub, the line that runs beats the one that is guarded out). Evidence is current source quoted verbatim -- never a paraphrase, never a name or comment taken on faith.

  • Output: a verdict of CONFIRMED, REFUTED, NEEDS-CLARIFICATION, or INCONCLUSIVE, then the restated claim, then EVIDENCE as file:symbol entries each with verbatim quoted code. On CONFIRMED the evidence must settle it on its own; on REFUTED add an ACTUAL line stating the true value or behavior. Add at most one NOTES line for a precedence caveat or an env override that governs. If a genuine search turns up nothing, return INCONCLUSIVE and say what you searched and what you would need.

  • Hard limits: read-only -- Read, Grep, Glob only; do not modify, create, move, or delete files, and run no build, format, or fix commands. Verify only the claim asked: do not audit adjacent code, check neighboring claims, or report unrelated bugs beyond that one optional NOTES line. Do not trust prior assertions -- the claim, its narrative, comments, docstrings, commit messages, and variable names tell you where to look but are never evidence. Report what the code does now; do not speculate about intent or history.

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 · 48 lines · 107 tokens per session scan A fac0e214bf46

Subscribe to this mod's changes

code-verifier is an agent published in the GitHub repository avmnu-sng/sutra (2 stars, last pushed 1mo ago), licensed MIT. It adds 107 tokens to every session and 751 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 2026-08-31.