Team playbooks and outcome insights

Open a workspace approval in Agent Team Inbox and change one or more server-declared fields. Enable Save as a Team example before choosing Approve and resume. The option is off by default and appears only after a field changes.

• Codelit stores privacy-safe field names, approval-step position, workflow version, and deterministic guidance.
• Original values, corrected values, prompts, source excerpts, identities, secrets, provider IDs, and external resource IDs are not copied.
• Every example is excluded from model training.
• The original approval record and its digest remain separate and immutable.

An active example can add one short, value-free reminder to the matching step of future hosted runs. Codelit retrieves examples only for the same deployment, workflow-version receipt, and step ID. The reminder tells the agent which fields deserve review and explicitly forbids inventing missing values.

• A changed workflow version receives no guidance from the older example.
• Removing an example immediately excludes it from future retrieval.
• Restoring it re-enables the same immutable example; it does not create a new version.
• If playbook retrieval is disabled operationally, hosted runs continue with their existing workflow behavior.

Keep Pin as a golden eval enabled when the saved correction represents a contract worth protecting. Codelit checks the exact workflow receipt for the reviewed approval boundary and step count. This evaluation makes no model or provider call and records the pinned example version with its result.

Open Team Insights from Agent Team Inbox. Outcomes groups stored hosted runs by Agent Team and immutable workflow version so changes in completion and intervention behavior remain comparable.

• Completed outcomes count terminal runs that finished successfully.
• Intervention rate counts runs with a recorded approval boundary.
• Failure rate counts failed or held runs among terminal runs.
• Recovery rate counts completed runs with bounded retry or recovery evidence.
• Actual model cost sums the stored approximate cost of observed runs.

For each Agent Team, enter how many manual minutes one completed outcome would normally take. Codelit multiplies that owner-supplied value by measured completed outcomes and labels the result Assumption everywhere it appears.

• No time saved value appears until an editor enters one.
• Clearing the assumption removes the estimate without changing measured history.
• The report shows how many Teams have assumptions so partial coverage stays visible.
• Codelit does not infer labor cost, revenue, or avoided work from model output.

Choose Remove on an example to exclude it from future runs and active evals while preserving a reversible audit state. Delete workspace removes its playbook examples and assumptions with the deployment. Delete account & data removes every owned example, assumption, approval, run, audit record, and identity.