Make an Agent Workflow Film from Reviewed Tasks and a Real Result

What does Agent Workflow Film show—and what must you supply?

Agent Workflow Film v1.1.0 turns authored tasks, dependency links, review declarations and a permitted result screenshot into a local review room, native editable PowerPoint and optional 28-second MP4s. The package costs USD 4.99 individually. It does not run an agent, execute the displayed command or measure productivity; supply actual execution evidence separately when that is the claim you want to make.

Start with python3 scripts/start.py for the standard-library visual editor on 127.0.0.1. No API key or package installation is needed to edit/export source. Keep its terminal open and stop it with Ctrl+C. Generating PowerPoint or video still requires the pinned Python dependencies.

Write task claims that your reviewed sources actually support

Read references/input-and-review.md. Supply product, display command, CTA, 1–12 source records and 2–8 tasks. Each source has an ID, location, observation and reviewed boolean. Each task has an ID, status, integer planned minutes, dependencies, source IDs, result and acceptance condition. A source location is authored text; the builder does not fetch or verify it.

Keep dependencies acyclic. A reviewed task must reference reviewed sources and reviewed prerequisite tasks. Planned, running and blocked work remain visible. The rightsConfirmed field is your permission declaration, not independent proof of image rights. The bundled screenshot is a dated Skillrove homepage capture used in a fictional practice story; it is not an agent execution recording.

Run the practice source before asking for a polished film

Run these commands from the complete installed skill into a new output directory. Open agent-room/index.html and workbench.html. Without --render the delivery explicitly has no rendered movie. Inspect tasks.csv, project.json and input-evidence.json before interpreting any status or duration. On Windows use .venv\Scripts\python.exe.

python3 -m venv .venv
.venv/bin/python -m pip install -r requirements.txt
.venv/bin/python scripts/build-film.py --input examples/project.json --output ./agent-room

Edit a real input and check the rebuilt evidence ledger

In Edit inputs, change the first task from 12 to 15 planned minutes and inspect Result preview. Export the source bundle, keep its image files beside project.json and rebuild into a fresh folder. input-evidence.json must report 48 planned minutes while reviewedTasks remains 2 and liveAgentExecution remains false. Changing an estimate does not change evidence of completion.

Try an unreviewed source for a reviewed task, a cycle or an incomplete prerequisite in separate inputs. Generation should fail before creating a delivery. Do not mark a task reviewed just to make a progress display complete. Review the actual result and acceptance condition first.

Choose a real local export and distinguish it from a recording

Add --render --formats landscape --edge 720 --fps 24 and choose a fresh output folder for one proof film. After review, request portrait and square if needed. The six source-derived scenes total 28 seconds. Watch the actual MP4, inspect captions and each crop, then check the manifest. Font coverage or overflow may stop an export; do not weaken those checks to deliver unreadable text.

PPTX text/cards/notes are editable native objects; a screenshot stays an image. --edits changes deck wording, not task evidence. Browser source edits require rebuilding before existing MP4/PPTX downloads change. Deliver the reviewed source, screenshot notices, task ledger, timeline, captions and integrity manifest. Label the film as a product/workflow explanation unless you separately captured a real operation.

Use an incident timeline when the task is reconstructing an actual event sequence

Check multilingual announcement claims before a language launch film