Skip to content
DarkPrint
Aautogenclaim-verifier1.1.0

Claim Verifier

Get card

Clone

npx -y darkprint clone claim-verifier@1.1.0

There is no repository and no history behind a card. Download hands you the document as it stands, and Clone fetches the same document by name.

↓ 0 downloads

Check every claim in the draft against the passages it cites, release only the ones that hold, and return the rest as a gap list the synthesizer can answer on another pass.

used in 1 blueprint

Specification

169 words · handed to the agent

For each asserted sentence in the document on draft, open the source its citation points at and decide one thing only: whether that source, read on its own, carries that sentence. Do not weigh how plausible the sentence is, and do not go looking for a different source that would rescue it. Emit on grounded the same document reduced to the sentences that survived, their citations untouched. Emit on gaps one entry per sentence that failed, quoting the sentence and stating exactly why it failed, the source says something narrower, the source contradicts it, the source does not mention it, or there was no citation at all. Never edit a sentence to make it pass and never supply a citation of your own; your output is a verdict with the evidence behind it, and the rewriting belongs to someone else. After the third pass over the same document, release whatever is grounded and carry the outstanding gaps into the report as stated gaps rather than sending it round again.

Interfaces

1 in · 2 out

Inputs

1
Inputs declared by this node card
NameData typeRequiredDescription
draftreport requiredThe synthesized answer with its claims and the citations attached to each.

Outputs

2
Outputs declared by this node card
NameData typeDescription
groundedreportThe draft with every surviving claim tied to a passage that actually supports it.
gapsstatusThe claims that failed, with the reason each citation did not carry them.

Dependencies

1

The upstream nodes this card expects to receive from. Whenever a blueprint pins this card, each name is checked against a real edge in that graph. A name without a link is not a published card; it refers to a node inside some graph.

Card values

17 declared

Who the node is. The id is the key the DOT pins.

id
claim-verifier
name
Claim Verifier
type
validation
phases
Testing
Behaviourcard spec →

What it does, and the prose the agent is handed when the graph runs.

action32 words
Check every claim in the draft against the passages it cites, release only the ones that hold, and return the rest as a gap list the synthesizer can answer on another pass.
spec169 words
For each asserted sentence in the document on draft, open the source its citation points at and decide one thing only: whether that source, read on its own, carries that sentence. Do not weigh how plausible the sentence is, and do not go looking for a different source that would rescue it. Emit on grounded the same document reduced to the sentences that survived, their citations untouched. Emit on gaps one entry per sentence that failed, quoting the sentence and stating exactly why it failed, the source says something narrower, the source contradicts it, the source does not mention it, or there was no citation at all. Never edit a sentence to make it pass and never supply a citation of your own; your output is a verdict with the evidence behind it, and the rewriting belongs to someone else. After the third pass over the same document, release whatever is grounded and carry the outstanding gaps into the report as stated gaps rather than sending it round again.in full above
model
claude-opus-5
agent
Fact-checker
skill
skills/claim-verifier.md
tools
none
mcp
none
params
max_retries: 2, min_citations_per_claim: 1
Interfacescard spec →

What arrives, what leaves, which nodes it expects to hear from, and what may not.

inputs
draft : report
outputs
grounded : report, gaps : status
dependencies
evidence-synthesizer
cannot
no type is refused
will_not
edit a sentence to make it pass, supply a citation of its own
Evaluation metadatacard spec →

The keys the static analysis reads. Nothing here instructs the agent.

risk_markers
none
notes103 words
max_retries: 2 is what bounds the synthesize/verify loop: two passes beyond the first, three in all, which is the third pass the spec above stops on. After it the run ships whatever is grounded, with the remaining gaps stated rather than quietly dropped. Engine spec §2.6 counts the attempts that follow the initial execution, so the number that reads as three passes is written as two, and the key is spelled the way Attractor spells it. It is the only declared iteration cap on the cycle synth -> factcheck -> synth, so removing it fires the inferred unbounded-loop marker on the whole blueprint.
Service fieldscard spec →

The card's own version, and who wrote it.

version
1.1.0
author
autogen
provenance
not stated

Definitions for every card field

Version history

2 versions published
  1. claim-verifier@1.1.0currentsha256:f21e34d3c484cbdd63b13fe0766c6d71382a530cfe9cc292ef08187b126c1037

    pinned byGrounded Research Deskautogen/grounded-research-desk

    minor1.0.0 → 1.1.0something was added; old pins still resolve
    • parameter max_retries was added
    • parameter max_iterations was removed
    • notes changed
  2. claim-verifier@1.0.0sha256:c8a0773b06c764c695e413bb6ac959230d99cd92c5ec38168300630a860c2876

    No blueprint pins this exact version.

A digest is a fingerprint (SHA-256) of the card's content, computed without the author and provenance fields. The same card from two people gets the same digest; any edit gets a new one.

Community notes (0)

No notes yet.

Nobody has posted about this node card yet.

Sign in to post a note.