The Trial flag: one session through a Run's code, one JSON line #1001

Parent

#977

What to build

A hidden flag on thirdshift that runs one agent session the way a Run does: through the same session code, with the Factory skills, the Session prompt for its session kind, Model evidence, review reports, and a Resume for killed background work. It runs in a prepared repository, on the Harness, Model and Effort it's given.

• Input: a versioned JSON request naming the repository, session kind, inputs, Harness, Model, Effort and output folder.
• Output: one versioned JSON line with the outcome, the timings, any void cause, the session's ID, and where its reports and log are.
• Where it writes: its log and reports go into the output folder, outside the factory's logs folder. It writes no Command record, so the Inbox, the Usual ranges and Silent ignore it.
• Versions: a build refuses a request version it doesn't know. Both Arms' builds need the flag, so the first Experiment's Baseline is the first release that ships it.
• Session kinds: it starts with the Spec review. The code-review suite and its Code review session #1016 adds the Code review.

Acceptance criteria

• With a fake Harness on a prepared fixture repository, the flag runs a Spec review session and prints one JSON line with every field.
• Nothing is written under the factory's logs folder, and no Command record is made.
• An unknown request version is refused with a clear message.
• The request and the line are documented and versioned.