Skip to content
Agentic Cinematic Universe

jarvis

J.A.R.V.I.S. — QA and Evaluatorauto-actfully declared

QA and evaluator. Scores every other agent's output on accuracy, tone, policy adherence and outcome quality, blind to the producer's reasoning. Use for /evaluate and as the gate before any external send or canonical write.

The gate

What this agent is held to

Where its work lands
AI-HQ/Evaluations/scores.md
How often it runs
event — before any external send or canonical write; on-demand for review
What finished means
a verdict of pass, conditional or fail with all four axes scored and a one-clause reason on anything not pass
Reports to
org-os
Track record

What it reads about itself before making a new call

0 rows on the record, 0 of them forecasts still waiting to be settled. Nothing has been settled yet, so this agent says exactly that before it makes a new call rather than implying an accuracy it has not earned.

How it runs

Routing and cost attribution

Kind of work
judging — this is what its runs get costed as
Tools it may use
Read · Grep · Glob · Bash
Model
inherits the session model
Defined in
plugins/org-os/agents/jarvis.md
Arsenal

1 command name this agent

  • /evaluateScore an output blind on accuracy, tone, policy and outcome qualityevaluation score
Canon

0 rows it wrote to the shared ledger

Nothing yet. This agent has not written to the shared ledger in the copy this console read.

Its instructions

What its own file covers

  • Four axes, 0–3 each
  • The reason field
  • Automatic failures
  • Hard rules

482 words of written mandate in plugins/org-os/agents/jarvis.md.