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.
git clone --depth 1 https://github.com/tznthou/claude-prismWrote 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/commands/tznthou/claude-prism/pi-fact-check)<a href="https://agentmods.dev/commands/tznthou/claude-prism/pi-fact-check"><img src="https://agentmods.dev/badge/commands/tznthou/claude-prism/pi-fact-check/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/commands/tznthou/claude-prism/pi-fact-check"><img src="https://agentmods.dev/badge/commands/tznthou/claude-prism/pi-fact-check.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 | $0.00014 | $0.02638 |
| Opus 5 | $0.00007 | $0.01319 |
| Sonnet 5 | $0.00003 | $0.00528 |
| Haiku 4.5 | $0.00001 | $0.00264 |
Grade A, and why
pi-fact-check 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 12d 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 — 221 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Fact-Check via Cross-Provider Verification
Verify factual claims using Gemini (Google search) for source discovery and Claude for cross-verification. Two providers, two roles: Gemini finds evidence, Claude judges it.
Execution
Bash invocation rules (v0.12.3+): Call
call-codex.sh/call-gemini.shin foreground synchronous mode withtimeout: 600000(Bash tool's 10-minute ceiling). Do not use&, do not setrun_in_background: true, do not usenohup— Claude Code 2026-04+ has an auto-background child-lifecycle regression that silently kills the process (output file stays 0 bytes). Userun_in_background: true+ BashOutput polling only as an exception when the call is genuinely expected to exceed 10 minutes.
1. Obtain content
Use $ARGUMENTS as the content to verify.
- File path: Read the file and use its content
- URL: Fetch with WebFetch
- Conversation reference: Claude extracts relevant content from history
- Inline text: Use directly
- Empty: Ask what to fact-check
2. Extract claims (Claude)
Scan the content. Identify every factual claim — statements that can be verified against external sources (data, events, quotes, attributions).
Skip opinions, predictions, and vague assertions ("many believe...").
For each claim, assess: how important is it if this is wrong? Focus effort on high-impact claims. Flag anything suspicious — too-precise numbers, vague sourcing, absolute language.
Output a numbered list of claims to verify, each with the exact text and suggested search terms. Prefix high-impact claims (the ones whose being wrong matters most) with [HIGH-IMPACT] — this marker travels into the Gemini batch prompt and directs deeper search effort.
If no verifiable claims found, report and stop.
3. Source search (dual-track)
Launch both tracks simultaneously in a single response to eliminate dead-wait time:
┌─ Track A: Gemini (search grounding) ─┐
Claims ──split──→│ │──merge──→ Step 4
└─ Track B: WebSearch (reliable URLs) ─┘
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.
- 12d ago First seen · 221 lines · 14 tokens per session scan A 74a3a76139a0
pi-fact-check is a command published in the GitHub repository tznthou/claude-prism (10 stars, last pushed yesterday), licensed MIT. It adds 14 tokens to every session and 2,638 once invoked, about $0.0001 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.
Other commands, from other repositories
conventions-check
Check AGENTS.md, CLAUDE.md, and the projected agent artifacts (skills, subagents, commands) against the repo's conventions.
greptile-review
Execute the full Greptile review cycle on a PR, looping until score meets threshold or circuit breaker triggers.
register-plugin-asset
A project command for adding a distributable command, agent, or agent skill to a plugin manifest, the file that lists what a plugin provides.
challenge
Adversarial review — run pre-mortem, war-game, and logic-torturing analysis on current changes.
logic-diff
Check two code versions for semantic equivalence — use after a refactor or rewrite.
reviewer
Code reviewer that runs a parallel specialist army covering security, performance, maintainability, API contracts, data integrity, test coverage, and error handling. Trigger on code review, review, PR review, pull request, or review army.