Fictional meeting transcript
A small, public test fixture with confirmed work, an unassigned analytics task, one idea, and explicit approval limits.
Back to Test Records
Two prompt-contract runs recorded
Chief of Staff prepares an operating brief from a meeting
Two recorded Hermes runs separated priorities, work, decisions, and ideas; the second also exposed two unsupported additions
Material supplied
The Bot begins with disclosed inputs
The meeting transcript is fictional. It is designed to test whether the Bot preserves owners, dates, approvals, unassigned work, and ideas without inventing commitments.
Download file
Exact disclosed run prompt
The role, request, fictional transcript, and acceptance checks sent in both runs.
Recorded request
The request used for both recorded runs
Turn this fictional sample transcript into a one-page operating brief. Separate confirmed priorities, assigned work, blocked decisions, and ideas. Preserve every owner and date exactly; write “unassigned” when an owner is missing. Flag capacity conflicts. Draft a five-item meeting agenda and a follow-up message for review. Do not assign people, change dates, contact anyone, or invent commitments.
Conversation evidence
What the record currently contains
These are concise excerpts from two independent Hermes CLI runs. The complete final responses and sanitized run summary are linked below. Provider-internal reasoning and session identifiers are not published.
Person · excerpt
Turn this fictional sample transcript into a one-page operating brief. Preserve every owner and date exactly; write ‘unassigned’ when an owner is missing. Do not invent commitments.
Bot · excerpt
Run 1 kept analytics unassigned, the newsletter as an idea, publication pending Maya’s approval, and the budget unchanged. It passed the disclosed assertions.
Bot · excerpt
Run 2 kept the core facts but added an unsupported ‘reply by end of day’ request and described Jordan as a ‘post-check owner.’ It did not pass every assertion.
Finished deliverable
What a person can inspect
Run 1 operating briefThe complete final response from the run that met the disclosed acceptance assertions.
Starter package
The files used to set up this Bot
These are the Bot profile and readable setup files. They are separate from any finished result produced during a run.
Inspect run 2 and its unsupported additions
The complete second final response is preserved rather than silently corrected.
Inspect the sanitized run summary
Hermes version, provider, model, timing, estimated cost, tokens, hashes, and assertion results. No session identifiers or secrets are included.
Download the Chief of Staff Hermes profile
Version 1.0.0 profile archive for Hermes Agent 0.20 or later.
Inspect the readable Chief of Staff files
ZIP package containing the same six readable source files.
Human decisions
The Bot prepares work; a person remains responsible
- 1
Confirm that the transcript is complete and accurate.
- 2
Assign the analytics work that the transcript leaves unassigned.
- 3
Approve priorities, commitments, and any follow-up before it is sent.
Evidence ledger
Separate checks for separate claims
Package, import, role-run, reproduction, and human-review status are reported independently.
Package checksPassed
The profile archive and ZIP contain the six declared files and match the readable copies.
Checked 2026-08-28Profile importNot run
The exact downloadable Chief of Staff profile was not imported for these runs.
Role runPartial record
Hermes Agent completed the disclosed role and request twice. One run passed the acceptance assertions; one added two unsupported details.
Checked 2026-09-02ReproductionNot run
Repeating the prompt without importing the exact profile is not a package reproduction.
Human technical reviewUnavailable
A human technical review is a possible long-term goal. Bot Cabinet cannot offer one at this time.