Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.armature.tech/llms.txt

Use this file to discover all available pages before exploring further.

Workflows are the core building block of Armature. Each workflow defines a tester prompt — the task an AI agent should accomplish — and a set of evaluation criteria that determine whether the run passed. Armature executes workflows on a schedule, records every tool call and trace event, and surfaces the results in the dashboard so you can catch regressions the moment they happen.

​
What a workflow contains

A workflow packages three things together: the prompt given to the tester agent, the evaluation criteria the judge model uses to score the run, and a schedule that controls when runs fire automatically. When a run completes, the judge evaluates each criterion independently and produces a per-criterion verdict, a roll-up status, and a full tool-call trace you can inspect. Every edit you make to a workflow creates a new immutable version. Armature keeps the full version history so you can trace exactly which prompt or criteria change coincided with a regression.

​
Workflow states

A workflow is either Active or Paused.
  • Active — the workflow runs automatically on its configured schedule and appears in coverage and health reports.
  • Paused — scheduled runs are suspended, but the workflow and all its past run data are preserved. You can resume it at any time by toggling it back to Active in the editor.
Pausing is useful during planned downtime or when you are iterating on a new version of the workflow and want to avoid noisy failures.

​
The workflow list

The Workflows page shows a table with one row per workflow. Each row displays:
ColumnWhat it shows
WorkflowName and description
ScheduleHuman-readable schedule label (e.g. “Hourly”, “Manual”)
Pass rate (7d)Percentage of runs that passed in the last seven days, color-coded green / yellow / red
Last runRelative timestamp of the most recent execution
StatusActive or Paused
Use the segmented control at the top of the page to filter to All, Active, or Paused workflows. The search box filters by workflow name or description — useful when you have many workflows across multiple MCP servers.

​
Finding and navigating workflows

From the workflow list you can:
  • Click Run history on any row to jump to that workflow’s runs filtered in the Runs page.
  • Click Edit to open the workflow editor, where you can update the prompt, criteria, schedule, and models.
  • Click New workflow to create a workflow from scratch.
The 7-day pass rate turns yellow when it falls below 95% and red when it falls below 80%. A healthy workflow stays green. If you see yellow or red, open the run history to identify which criteria are failing.

​
Workflow versions

Every time you save changes to a workflow, Armature creates a new immutable workflow version. The current version is what the tester agent runs. The version history is visible in the History tab of the workflow editor and is accessible via the MCP repair API for diffing and regression analysis.
If you need to experiment with a major prompt change without affecting production runs, pause the workflow first, make your edits, and re-enable it once you are satisfied with a test run.

​
Learn more

Authoring workflows

How to write effective prompts and evaluation criteria that produce reliable, deterministic results.

Scheduling

Choose the right schedule for your workflow — from manual-only to fully automated cron runs.

Run results

Understand pass, partial, fail, and error statuses. Inspect traces, compare runs, and diagnose failures.

MCP API overview

Use the Armature MCP repair API to propose patches, trigger runs, and compare results from your own agents.