Use the installed taskfilm CLI. Read its README for installation and run
taskfilm doctor and taskfilm capture check before browser capture. This skill
is self-contained and does not require the private monorepo or fleet tools.
Choose the story treatment from the user's goal: feature_demo for one
repeatable task, saas_launch for problem/workflow/payoff and a supplied CTA,
github_review for a pinned OSS source with fit, limitation and attribution.
taskfilm init demo supplies an original working app, capture script and sample
request/story files. Adapt its inputs instead of inventing unsupported flags.
Read taskfilm story --help and taskfilm capture --help for the actual contracts.
Keep before/action/result explicit. For a lesson, keep real interactions at
natural speed and hold the visible result long enough to read. Promos can cut
idle time, with a useful result still on screen. Use taskfilm repo-pack and
story --repo-pack for source-based OSS assertions; the source pack does not
prove runtime behaviour.
For the original Taskfilm OSS sample, init includes oss.json and
prepare-oss-review.py. Bind its request to the actual pinned public release
README using the recipe in the README, then use story --repo-pack. The helper
refuses README bytes that differ from the source pack. Other repositories need
their own source claims and real task actions.
Record with taskfilm capture rec SCRIPT --out-dir takes --no-burn-captions.
Use the exact take path it reports. taskfilm edit TAKE --no-retime prepares
the lesson edit. taskfilm compose TAKE --out PROJECT or taskfilm reel TAKE --out PROJECT exports an editable composition from real footage and beat
labels. The story brief is direction to apply and inspect, not automatically
executed by the composition command. taskfilm render PROJECT --output FILE
checks and renders through the pinned HyperFrames owner.
For a directed motion promo, use the existing product-film owner: the sample
motion-capture.json records a real 4K take, and motion-spec.json supplies the
brand, copy, measured regions and action anchors. Copy the spec to spec.json
beside the film, then run taskfilm motion FILM --take EXACT-TAKE --format landscape
or portrait. Read taskfilm motion --help. Its optional NumPy/SciPy dependencies
are installed with the motion extra. The output under FILM/cuts/FORMAT
retains an edit-decision record; rendering stays with taskfilm render.
For pure motion graphics, init also supplies motion-lab/: four original
editable HyperFrames scenes with masked type, SVG drawing/interpolation and
perspective. Edit the scene files and their paused GSAP timelines, then render
that project. It is a silent animation study. Do not describe it as an Adobe
After Effects project importer.
taskfilm package --showcase MANIFEST --out DIRECTORY packages finished,
hash-bound movies and distribution downloads as a portable static release
preview. It probes movie metadata and retains hashes; it does not create or
approve a film. package --evidence retains the original take-companion contract.
Keep plans, capture success, verified product result and final-video review distinct in your report. Inspect the result in the captured task, inspect delivery-size video frames, and listen to complete audio when supplied. Never treat a UI replica as proof. Use supplied/currently owned accounts and media. Keep cookies and browser profiles out of examples and packages. Publish only when the user has authorized the concrete destination/output.