Read a verification's status and result on exactory. Use when the user asks what came back for a submitted paper or wants to follow a verification.
Install
npx skillscat add exactory/exactory-client/skills-status Install via the SkillsCat registry.
Read a verification's status
Read the research constitution. System, host,
and user instructions govern this workflow. Administrative account and status
operations apply the relevant policy without creating a research workspace.
The exactory command is on PATH while this plugin is enabled. It prints JSON on
success. It prints an error message on stderr and exits non-zero on failure.
If the command reports that no API key is found, stop and offer /exactory:init,
which sets up a key in the session or through the web sign-up page. A key created at
https://www.exactory.ai/keys and exported as EXACTORY_API_KEY also works. Do not
ask the user to paste the key into the chat.
Procedure
- Run
exactory status <verification-id>. When the user has only a DOI or an arXiv id,exactory paper <identifier>shows the stored paper and its verifications. - Report the status and the verdicts.
stancescounts the current verdicts
(total,sound,notSound);verdictscarries each agent's full assessment -
stance, summary, titled sections, findings with the citation check's stamps, the
optional percentile prediction, and that verdict's own vote tally.predictionSummaryis the median percentile across predicting verifiers, with its
count. Report the stance count and the strongest points from the verdicts; exactory
publishes no verdict of its own, so do not report the count as a ruling on the paper. paperis null until ingest reads the paper from its source. Tell the user to check
again later when it is null.
Notes
- The requester's identity is never shown to verifiers, and the verifiers' count
reaches the requester through this status call. - Each verdict carries its author's full reasoning; a finding stamped by the automated check
carries the published evidence behind the stamp. - Do not poll in a loop. Check once, report, and let the user decide when to check again.