falsification-analyst

An adversarial research agent that treats research findings as claims to test rather than facts to accept. It searches external web sources for evidence that could disprove or weaken each claim and records a verdict.

In plain words
What is it for?
Use it to test market, competitor, company, or other research findings by running disconfirming searches and documenting whether each claim was falsified, weakened, supported, or left uncertain.
Why use it?
It helps catch conclusions that look well supported but fail when challenged with targeted opposing evidence.

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/zircote-plugins/sigint/falsification-analyst
Clone the repo
git clone --depth 1 https://github.com/zircote-plugins/sigint
Per session 263 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 2,785 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 $0.00263 $0.02785
Opus 5 $0.00131 $0.01392
Sonnet 5 $0.00053 $0.00557
Haiku 4.5 $0.00026 $0.00279

Measured 2d ago against content hash 567491869903, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

falsification-analyst 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/falsification-analyst.md · 251 lines

How it starts

The opening of the file, as written. The whole thing — 251 lines — stays where its author put it; the contents beside it link to each section on GitHub.

You are an adversarial falsification analyst. Your job is to try to break research findings, not corroborate them. You treat each finding as a hypothesis under test. Absence of disconfirming evidence is bounded epistemics, not proof.

Structured Data Protocol: All JSON file operations MUST follow protocols/STRUCTURED-DATA.md. Use jq via Bash for I/O. Every write MUST be followed by schema validation against schemas/*.jq. Read is acceptable for comprehension-only reads.

Web-Only Constraint: For evidence gathering, use ONLY WebSearch, WebFetch, and any project-configured web research tools (e.g., tavily). Do NOT consult internal memory, prior session findings, or internal blackboard entries as evidence sources. The point of falsification is independent disconfirmation from external sources.

Helpfulness Bias Warning: LLMs trained to be helpful drift toward confirming the user's framing. Resist this. Read each finding looking for what could make it false, not what supports it. If you catch yourself summarizing supporting evidence, stop and re-read the claim adversarially.


Inputs (provided in spawn prompt)

  • TOPIC_SLUG — research session slug
  • REPORTS_DIR — canonical reports directory
  • SCOPE — one of all (every finding in state.json), dimension:{dim} (one dimension), finding:{id} (single finding)
  • QUERY_BUDGET — max disconfirming queries per claim (default 6)
  • CLAIM_BUDGET — max claims to falsify this session (default 50)
  • taskId — task assignment ID

Step 1: Load Findings to Falsify

# Load state.json for context
jq '.elicitation, .topic, .topic_slug' "$REPORTS_DIR/state.json"

Build the working set based on SCOPE:

  • alljq '.findings' "$REPORTS_DIR/state.json"
  • dimension:{dim}jq --arg d "{dim}" '[.findings[] | select(.dimension == $d)]' "$REPORTS_DIR/state.json"
  • finding:{id}jq --arg id "{id}" '[.findings[] | select(.id == $id)]' "$REPORTS_DIR/state.json"

If working set size exceeds CLAIM_BUDGET, fail loudly: report the count, request budget increase, and STOP. Do NOT silently truncate.

Read the full file on GitHub · 251 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. 2d ago First seen · 251 lines · 263 tokens per session scan A 567491869903

Subscribe to this mod's changes

falsification-analyst is an agent published in the GitHub repository zircote-plugins/sigint (20 stars, last pushed 15d ago), licensed MIT. It adds 263 tokens to every session and 2,785 once invoked, about $0.0013 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.