Independent critic loop: example workflow - Pathista help center

Research on language models points the same way. A model asked to judge can favour its own output, and models struggle to correct their own reasoning without external feedback. A critic that did not do the work, and that opens the files and runs the checks itself, is the remedy the template asks for.

Pathista does not run the critic. The template tells the assistant to hand the result to an independent reviewer, ideally a separate agent or session, and to record the verdict. The run's log shows what the assistant says the critic did and found.

• The goal and the bar come first: a short list of criteria that someone who did not take part can check, each tied to an artifact such as a file, a screenshot or a command.
• The work is split into parts small enough to be checked in one sitting.
• Each round the assistant builds and does not grade its own work; from the second round it fixes only the one gap the critic named.
• The critic returns one of two verdicts: the single largest gap with evidence, or a statement that the bar is met with evidence for each criterion.
• The loop ends when the bar is met or when the iteration counter reaches maxIterations, which is 3 unless you change it. In the second case the verdict says plainly that a gap is still open.

Add the template to your workflows and set the budget and the bar for your work. It suits anything you can state a bar for: a document, a feature, a design.

Words in code type, such as iteration, are the template's own variables: small yes/no values or counters that the assistant sets during the run and that decide which way out of a step is open.

The checklist of the step “Critique”. The assistant may leave the step only when every item is closed with evidence:

• The critic worked in a fresh context (separate agent or session) and saw only the goal, the bar and the result
• The critic ran or opened the artifacts itself, and its verdict cites their output
• The verdict is either ONE largest gap with evidence (path, failing check or file:line) or an explicit statement that the bar is met
• The verdict names the bar criterion it relies on, and its text was quoted into a note
• The route taken matches the verdict and the iteration counter: gap found, meets the bar or round limit

Every step and checklist item is on the Gauntlet loop template page.

Open the template

• Claude Code release checklist: example workflow
• Customer interview synthesis: example workflow
• What is an MCP workflow?
• Concepts: the words Pathista uses.
• Connect an assistant: one link and one sign-in.
• All guides