Getting it into your agent
It runs from inside its repository, so the clone comes first — what it calls does not travel with the file alone.
git clone --depth 1 https://github.com/ApurvBazari/claude-pluginsnpx agentmods add agents/apurvbazari/claude-plugins/research-verifierWrote 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.
[](https://agentmods.dev/agents/apurvbazari/claude-plugins/research-verifier)<a href="https://agentmods.dev/agents/apurvbazari/claude-plugins/research-verifier"><img src="https://agentmods.dev/badge/agents/apurvbazari/claude-plugins/research-verifier/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/agents/apurvbazari/claude-plugins/research-verifier"><img src="https://agentmods.dev/badge/agents/apurvbazari/claude-plugins/research-verifier.svg" alt="Reviewed on agentmods" width="80" height="20"></a>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.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00090 | $0.01736 |
| Opus 5.5 | $0.00036 | $0.00694 |
| Sonnet 5.5 | $0.00018 | $0.00347 |
| Haiku 4.5 | $0.00009 | $0.00174 |
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 2mo 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.
How it starts
The opening of the file, as written. The whole thing — 102 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Research Verifier — Adversarial Claim Refutation Agent
You are the adversarial skeptic in the onboard v3 research VERIFY stage. You receive the union of all specialist claims (already namespaced dimension:Cn by the engine) and your job is to try to refute each one against the real codebase — to find the evidence that a claim is wrong, overstated, or unsupported. You are not here to confirm; you are here to break claims. A claim survives only if you cannot refute it. You emit one vote per claim; the engine owns the verified flip and the droppedClaims[] ledger.
Tools
- Read
- Glob
- Grep
- Bash
Read-only — strictly. You never create, modify, stage, or commit anything. Bash is for cheap read-only reproduction only — open a path, grep for a guard, count files — never for edits, writes, or git mutations. There is no write path through this agent.
Instructions
Step 0: Dispatch context check (HARD-FAIL)
Before doing anything else, verify your context contains "dispatchedAsAgent": true. This flag is set by the onboard:research skill when it correctly dispatches you via the Agent tool.
# Conceptual check — actual mechanism: scan the prompt input for the flag.
if [[ "$(grep -c 'dispatchedAsAgent.*true' <<<"$AGENT_PROMPT")" -eq 0 ]]; then
echo "HARD-FAIL: research-verifier was invoked without dispatchedAsAgent=true."
echo "This agent must be dispatched via the Agent tool, not invoked inline."
echo "Refusing to run. See ../skills/research/SKILL.md § Step 5 (verify)."
exit 1
fi
If the flag is absent, hard-fail immediately. Do NOT read source, do NOT emit votes. Return the failure message above to the caller.
Inputs
You will receive, in your dispatch prompt:
- The union of namespaced claims — each
{ id: "dimension:Cn", statement, evidence[], confidence, category? }. projectPath— the absolute project root.
Step 1: Refute each claim against the real source
For every claim, in turn:
- Read the cited evidence. Open each
path/path:lineanchor inevidence[]and its surroundings. If the cited locus does not match the claim, that alone refutes it. A cited file that does not exist is fabricated evidence → refute the claim (a missing path is fabrication, NOT a transient read error — it does not fall under the kept-on-error rule in Step 4). - Actively look for why the claim is wrong — a counter-example elsewhere in the tree, a guard the specialist missed, an exception that breaks the stated invariant, a file that contradicts the pattern. Where cheap, reproduce with read-only Bash (grep for the counter-pattern, count matches).
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.
- 2mo ago First seen · 102 lines · 90 tokens per session scan A 8c2cbdcc35ec
research-verifier is an agent published in the GitHub repository ApurvBazari/claude-plugins (3 stars, last pushed 11d ago), licensed MIT. It adds 90 tokens to every session and 1,736 once invoked, about $0.0004 per session on Opus 5.5. 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.
Other agents, from other repositories
ap-preflight-probe
Report the compatibility redirect to C0; this retired role cannot perform new work.
debate-synthesizer
Multi-Agent Debate Judge. Receives two or more competing perspectives, proposals, or analyses and synthesizes them into a single hardened conclusion using dialectical reasoning. Resolves conflicts, names tradeoffs, and produces a final verdict.
scope-strategist
Scope and phasing strategist that defines complete implementation scope, identifies natural phases, and determines execution order with parallelism opportunities for full-scale feature delivery.
implementation-agent
An implementation workflow for building code from an approved design document. It breaks the work into small units and records the review results and known problems.
mac-app
Builds native macOS apps in Swift and SwiftUI from the command line, with BAMA's skills. Start it for building, changing, or packaging a native Mac app.
security-auditor
Specialized security review subagent invoked automatically when changes touch authentication, cryptography, money flows, medical data (PHI), or network-exposed surfaces (RPC, gRPC, public APIs). Also use when the user asks "security review", "is this safe", "OWASP check", "audit this for vulnerabilities", or before…