Skip to content

What you get

Compiling a demonstration produces two things: a bundle you deploy and, on every replay, a run report you can read and audit.

The bundle

A workflow bundle is a self-contained directory, the deployable artifact: version it, review it, ship it. It holds the compiled workflow program, the per-step evidence the replayer needs to re-find each target, and the policy metadata lint and certify read.

Because the bundle is the unit of deployment, everything about a workflow is auditable before it runs: which steps write, which clicks are identity armed, which postconditions each step asserts. See The bundle format for the full layout.

The run report

Every replay writes a timestamped directory under runs/ with an illustrated REPORT.md (human-readable) and a report.json (machine-readable). Review or sanitize either before it crosses a boundary. The report is the audit trail. For each step it records:

  • Resolution: which rung of the ladder resolved the target (template, OCR, geometry, or a grounding model), and whether a heal was applied.
  • Identity: whether the step was identity armed, and what the pre-click check verified or refused.
  • Postconditions: which assertions passed, and any effect verified against the system of record.
  • Outcome: success, or a halt naming the violated expectation.

The report also states, in plain language, how many click steps were identity armed (for example "4 of 12 click steps identity-armed") and lists any unarmed steps with the reason.

Emit for other agents

A compiled bundle is not only a replay target. Emit it as an Agent Skill or a standalone MCP server so other agents can invoke the workflow as a tool:

openadapt flow emit-skill bundle --out skills/
openadapt flow emit-mcp   bundle --out server.py

Privacy of the artifacts

The REPORT.md and console logs can be processed by the PHI/PII sanitizer on the persist/log path (Presidio-backed, via the optional privacy extra). The compiled bundle and report.json keep literal identifiers on purpose: they are the identity check and the audit trail, protected by a documented boundary rather than by redaction. Detector misses remain possible, so review sanitized output before egress. See Deploy on-prem.