A field guide for data-portable Muse work

Make Muse’s invisible work legible.

FedPromptly is a practical set of prompts, templates, and habits for seeing what Muse by Meta actually did: action by action, byte by byte, with uncertainty left on the page.

Read the method Copy the prompt

Put a ruler in the room before you send an agent into the world.

Most “usage” reports collapse a multi-step job into one vague number. This project refuses that shortcut. Define the unit. Capture the evidence. Label the gap.

Count actions, not vibes.

A search, page load, upload, form fill, model call, and download are separate events. Each receives an ID and a timestamp.

Separate the instruments.

Network bytes and model tokens are different measurements. Keep input, output, reasoning, cache reads, and latency in their own lane.

Make uncertainty visible.

Measured, reported, estimated, and unavailable are not synonyms. A useful ledger tells the reader which one they are looking at.

The page is a small control room.

Use the same vocabulary in the setup prompt, the interim table, the final report, and the conversation around it.

ACTION
what happened?

EVIDENCE
what can we prove?

REPORT
what should change?

From prompt to operating rhythm.

The included prompt starts in Data Usage Audit Mode. It asks the agent to make a ledger entry before moving to the next action, then surfaces a compact interim table after every five actions.

Ask before sign-in, upload, download limits, or form submission.
Record category and size instead of retaining sensitive content.
Stop before an action expected to exceed the agreed data threshold.
End with totals by action type, service, and evidence quality.

$ cat prompts/usage-tracker-setup.md
# copy into a new agent project before the task

A ledger you can actually read.

Here is the shape of the final report. The rows are intentionally plain; the point is to make the evidence hard to hide.

Action	Type	Target	Sent	Received	Total	Method
A-001	Search	Search engine	1.8 KB	420 KB	421.8 KB	Estimated
A-002	Open page	example.com	0 KB	2.4 MB	2.4 MB	Measured
A-003	Summarize	Muse model	12 KB prompt	Token usage separate	N/A	Reported
A-004	Download	report.pdf	0 KB	8.1 MB	8.1 MB	Measured

Operating rule: if the agent cannot see the network measurement, the right output is Unavailable — not a made-up KB or MB figure.

Keep the work open

Good prompts should travel. Help them travel further.

Support the public toolkit.

Contribute to the documentation, share the prompt with someone running multi-step Muse work, or help fund careful maintenance.

See support options