CtrlK
BlogDocsLog inGet started
Tessl Logo
PluginPublic

jbaruch/speaker-toolkit

Twelve skills for conference speakers covering rhetoric analysis, talk authoring, deck building and export, CFP submissions, publishing, illustrations, shownotes, screencast production, and vertical shorts, informed by a 113-entry Presentation Patterns taxonomy (83 observable: 64 patterns + 19 antipatterns; 30 unobservable: 21 patterns + 9 antipatterns).

Quality

92%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide
SecuritybySnyk

Medium

Suggest reviewing before use

Overview
Quality
Evals
Security
Files

shorts-vision-validation-authority.mdrules/

alwaysApply:
No
applyTo:
skills/shorts-producer/scripts/vision.swift, skills/shorts-producer/scripts/prepare-clip.py, skills/shorts-producer/scripts/verify-short.py — when changing how the shorts producer asks Apple Vision for faces or QR payloads
description:
Authority of record for the shorts producer's Apple Vision boundary under the Platform-Bound Untestable Carve-Out

Shorts Vision Validation Authority

Carve-Out Claimed

  • jbaruch/coding-policy: testing-standards Platform-Bound Untestable Carve-Out.
  • Qualifying condition: Apple's Vision framework is a proprietary macOS runtime. The suite that tests the shorts producer runs on Ubuntu, and the macOS lane runs only the artifact-platform contracts.
  • Scope: Apple Vision face and QR detection only. The word-timing layer is governed by rules/transcript-fetch-authority.md.

Covered Artifact

  • Exempt: skills/shorts-producer/scripts/vision.swift and its install mirror vision.swift.txt, in both modes, faces and qr. vision_script in skills/shorts-producer/scripts/shorts_plan.py copies whichever is present to a runnable name, and that copy is tested.
  • Exempt: the two calls that run it, detect_faces in skills/shorts-producer/scripts/prepare-clip.py and _scan_qr in skills/shorts-producer/scripts/verify-short.py. Each returns Vision's JSON rows; detect_faces only turns each frame's file name into its time.
  • Not exempt: everything that consumes the detections. Region filtering, crop size, smoothing, keyframes, the crop expression, caption alignment and the edit-budget refusal, caption timing, layout, rendering and the other verification checks are deterministic and tested.

Precondition 1 — CI-Runnable Pieces Are Extracted and Tested

  • tests/test_shorts_producer.py feeds synthetic face rows to the crop tracker and asserts region filtering, crop size, smoothing and the refusal when too few faces fall in the region.
  • The same file renders a short end to end from a synthetic recording, deck page, word timings and face rows, then runs every verification check against it, with the QR scan's payloads supplied in place of Vision's.

Precondition 2 — Manual Validation Procedure

Run on macOS against a real processed talk with a visible speaker and a shownotes link.

  1. Run the skill through Step 5 for one clip. Observe: faces.json has a face on the speaker in most frames, and contact.jpg shows red boxes on the speaker's face.
  2. Complete Steps 6 and 7. Observe: the rendered speaker panel keeps the speaker's face in frame for the whole clip.
  3. Run Step 8. Observe: qr_payloads holds the plan's qr_url and ok is true.
  4. Set the clip's qr_url to a different https link without re-rendering and rerun Step 8. Observe: it fails, naming the QR mismatch.

A pass requires all four observations.

Scope Limits

  • The carve-out covers only the Vision requests named above. It does not extend to ffmpeg, Pillow or qrcode, which CI installs and exercises.
  • Adding another exempt artifact requires naming it here with its own validation procedure.

README.md

tile.json