Reporting the outcome of an agent task: done, partial or blocked, with evidence

Goal

Deliver a final report whose status the reader can take at face value and act on, and which makes a partial result as useful as a complete one.

Prerequisites

A task with a stated goal and, ideally, an acceptance check agreed at the start. If none was agreed, the report states the check the agent applied.

Steps

1. Choose exactly one status. Done: the acceptance check passed and there is evidence. Partial: some of the scope is done with evidence and the rest is listed. Blocked: no further progress is possible without an input, a permission or a decision, and that input is named. Do not upgrade a partial to done because the remainder seems small.
2. Lead with the status and the one-sentence result, then the evidence: which checks ran, with their command and outcome (test counts, a status code, a diff summary). Evidence the reader can repeat is worth more than a description.
3. List what changed, by identifier: files, records, configuration keys, external objects. Separate reversible changes from ones that already had an external effect.
4. State every assumption the work rests on, at the point where it matters, together with what would change if the assumption is wrong.
5. For partial and blocked: list the remaining scope as concrete steps, and for each blocker the exact input that would unblock it (a decision between named options, a credential, an answer to a written question).
6. Report failures plainly with their output. A test that fails is reported as failing, not as "mostly passing"; a step that was skipped is reported as skipped, with the reason.
7. Keep the report free of narrative about the process unless the process produced a finding the reader needs (a surprising constraint, a wrong assumption in the brief). Dead ends go into the run log, not the report.

Expected result

A reader who sees only the report can decide whether to accept, continue or redirect the work, and can verify the "done" claim from the evidence lines alone. Partial results are handed over without loss because the remaining scope is already written down.

Limits and test basis

A proposed format, not a measured one. Its value depends on the acceptance check being stated early; a report can be honest about a vague goal only by saying that the goal was vague.