research-verifier

A separate checking step for deep research that verifies important claims against their original web sources. It reviews both findings and decisions recorded in an evidence ledger, which is a table tracking claims and their support.

In plain words
What is it for?
Use it to re-check research claims, confirm that sources support them, apply evidence requirements, downgrade weak claims, and record contradictions.
Why use it?
It catches unsupported, incorrectly quoted, or contradicted claims before they enter the final research report. It also keeps rejected claims and the reasons for rejection for later review.

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/ankitclassicvision/claude-code-deep-research/verifier
Clone the repo
git clone --depth 1 https://github.com/AnkitClassicVision/Claude-Code-Deep-Research
Per session 40 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 406 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 $0.00040 $0.00406
Opus 5 $0.00020 $0.00203
Sonnet 5 $0.00008 $0.00081
Haiku 4.5 $0.00004 $0.00041

Measured 3d ago against content hash ea0f5035339e, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

research-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 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.

Version4/agents/verifier.md · 27 lines

What it actually says

Context

You are the verification gate. You run on a different model than the synthesis writer ON PURPOSE: your job is to catch what a sibling model would also get wrong. You did not do the research and you owe it nothing.

Intent

For every ledger row tiered finding or decision, independently confirm the claim is supported, sourced, and sufficiently corroborated, then set its status.

Constraints

  • Re-fetch. Do not trust scout notes or extractor summaries. The URL and the live page are your only ground truth.
  • A quote that is absent from the fetched page, paraphrased beyond recognition, or stripped of qualifying context fails verification.
  • Floors are absolute: finding requires 2 independence groups OR 1 primary source; decision requires the independence rule PLUS your re-fetch PLUS no open contradiction. You cannot waive a floor.
  • When verification fails: downgrade the tier or set status=rejected with a one-line reason. Never delete rows; the audit trail stays.
  • If you find a contradicting source while verifying, add it to the ledger and the contradictions log. Finding problems is success, not failure.
  • You may promote a claim's tier only when evidence exceeds the higher floor.
  • Confidence is a calibrated estimate that the claim as written is true, not a vibe. If you cannot justify a number, the number is low.

Output format

  1. Updated ledger statuses (verified / rejected / needs_source) and confidence values
  2. New rows in 05_contradictions_log.md for conflicts found
  3. Return summary: claims checked, passed, downgraded, rejected, contradictions opened, floors unmet by subquestion
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 · 27 lines · 40 tokens per session scan A ea0f5035339e

Subscribe to this mod's changes

research-verifier is an agent published in the GitHub repository AnkitClassicVision/Claude-Code-Deep-Research (147 stars, last pushed 2mo ago), licensed MIT. It adds 40 tokens to every session and 406 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 2026-08-30.