verify-claim

Verifies user-supplied claims against codebase, documentation, and web sources with sourced verdicts.

Updated Jul 22, 2026
One-click install
npx skills add https://github.com/Trigent-Software-Pvt-Ltd/trigen-aidlc --skill verify-claim-trigent-software-pvt-ltd
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: verify-claim
Source: https://github.com/Trigent-Software-Pvt-Ltd/trigen-aidlc/tree/main/plugins/verify-claim/skills/verify-claim
Command: npx skills add https://github.com/Trigent-Software-Pvt-Ltd/trigen-aidlc --skill verify-claim-trigent-software-pvt-ltd

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? It determines whether a technical claim is true, false, or unverifiable by delegating investigation to a dedicated agent that gathers evidence from the codebase, documentation, and the web, returning a sourced verdict with a confidence score. ## Core Features & Use Cases - Delegated investigation: Spawns an isolated claim-verifier agent via the Task tool so evidence gathering does not pollute the main conversation context. - Structured verdict contract: Returns a fixed markdown block with Verdict, Confidence (0-100 with rubric bands), Evidence for, Evidence against, and Gaps sections. - Hallucination detection: Flags unverifiable claims and suggests possible sources of confusion, such as mistaking a --resume flag for an invented --mind-meld flag. - Use Case: Ask whether GitLab CI needs: and dependencies: keywords are interchangeable, and receive an Unclear verdict with citations showing conflicting behavior across CI versions. ## Quick Start Ask the assistant to verify a specific claim, for example: verify the claim that Claude Code's allowed-tools frontmatter restricts which tools a skill can invoke.

Frequently Asked Questions about verify-claim

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I verify a technical claim with Claude Code?

Invoke the skill with the claim as an argument, such as /verify-claim:verify-claim followed by the claim text. It delegates the investigation to a dedicated agent and returns a verdict with a confidence score and cited evidence.

How does claim fact-checking against a codebase work?

The skill spawns an isolated claim-verifier agent through the Task tool, which searches the codebase, documentation, and web for evidence. The agent returns findings in a fixed markdown format with evidence for, evidence against, and gaps sections.

What verdicts and confidence scores does claim verification return?

Verdicts range from Verified (90-100 confidence) through Likely true, Unclear, and Likely false, down to Hallucination / unverifiable (0-14). Each verdict includes citations from files, URLs, or command output supporting the conclusion.

Can claim verification detect hallucinated CLI flags or API features?

Yes, claims about nonexistent features receive a Hallucination / unverifiable verdict with low confidence. The output also includes a Possible source of confusion section explaining what real feature the claim may have been mistaken for.

What happens if the verification agent returns malformed output?

The skill re-invokes the agent once, requesting the missing section. If the second attempt still fails the output contract, it surfaces the raw agent output to the user with a note rather than silently discarding it.