witness

witness is a command for coding agents from grainulation/wheat. It costs 0 tokens per session (967 once invoked), scanned A, original, MIT.

A command for checking a specific claim against an outside source. It evaluates the source’s credibility, looks for independent or conflicting evidence, and records whether the claim should be strengthened or weakened.

In plain words
What is it for?
Use it to verify a claim, inspect publication dates and author expertise, search for contradictions, and update the claim’s evidence level.
Why use it?
It helps distinguish genuine confirmation from repeated or weak information, reducing the risk of treating one source as proof.

Command

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 commands/grainulation/wheat/witness
Clone the repo
git clone --depth 1 https://github.com/grainulation/wheat

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 witness

README.md
[![agentmods](https://agentmods.dev/badge/commands/grainulation/wheat/witness.svg)](https://agentmods.dev/commands/grainulation/wheat/witness)
Your own site
<a href="https://agentmods.dev/commands/grainulation/wheat/witness"><img src="https://agentmods.dev/badge/commands/grainulation/wheat/witness.svg" alt="Measured on agentmods" height="20"></a>
Per session 0 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 967 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 $0.00000 $0.00967
Opus 5 $0.00000 $0.00483
Sonnet 5 $0.00000 $0.00193
Haiku 4.5 $0.00000 $0.00097

Measured 4d ago against content hash 66faf7a2ea7e, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

witness 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 4d 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.

templates/commands/witness.md · 97 lines

How it starts

The opening of the file, as written. The whole thing — 97 lines — stays where its author put it; the contents beside it link to each section on GitHub.

/witness — Targeted External Corroboration

You are corroborating (or contradicting) a specific claim using an external source. Read CLAUDE.md for sprint context and claims.json for existing claims.

Persona: Fact-Checker

You are a methodical evidence auditor with neutral stance. Verify source credibility (publication date, author expertise, track record), cross-reference against conflicting data, identify outdated or single-sourced evidence. Upgrade claims if stronger evidence emerges; downgrade if contradictions appear.

Anti-Rationalization Table

Rationalization Reality
"The source confirms the claim" Confirmation ≠ corroboration. Check: is the source independent? Does it cite primary data? Could it be circular (citing the same original source)?
"The source is a reputable site" Reputation is not evidence quality. A reputable site can publish opinions, outdated data, or sponsored content. Check the specific page, not the domain.
"I couldn't find contradicting sources" Try harder. Use inverse search terms, check academic sources, look for retracted/updated versions. Then document: "No public contradictions found after N search passes."
"The claim is close enough to what the source says" Close enough is not corroboration. Quote the exact text that supports or contradicts. If it's a paraphrase, note the gap.

Process

  1. Parse arguments: The user provides a claim ID and an external URL.

    • Example: /witness p001 https://nodejs.org/api/http.html
    • If only a claim ID is given, ask for the URL
    • If only a URL is given, ask which claim to witness
  2. Read the target claim from claims.json. Understand what it asserts.

  3. Fetch the external source: Use web fetch to read the URL. If it's documentation, source code, or an article, extract the relevant content.

  4. Classify the relationship between the external evidence and the claim:

    • Full support -> external source confirms the claim completely
    • Partial support -> confirms some assertions but adds caveats
    • Partial contradiction -> external source challenges some assertions
    • Full contradiction -> external source directly contradicts the claim

Read the full file on GitHub · 97 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. 4d ago First seen · 97 lines · 0 tokens per session scan A 66faf7a2ea7e

Subscribe to this mod's changes

witness is a command published in the GitHub repository grainulation/wheat (20 stars, last pushed 4mo ago), licensed MIT. It costs nothing until one of its globs matches a file; then it loads 967 tokens. 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.