Content
88%Weight 40%Scale 1-5Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
A strongly engineered orchestration skill: every step is executable with concrete commands, explicit gates, error-recovery loops, and mode branching, with details correctly pushed to one-level-deep references. The two real defects are a missing bundle file (`sub-agents/frame-worker.md`) that Step 5's dispatch treats as required input, and duplicated orchestrator statements that cost tokens without adding information.
Suggestions
Add the missing `sub-agents/frame-worker.md` (or correct its path) — Step 5's packet builder concatenates it verbatim into `_role.md`, so the current citation points at a file that does not exist in the bundle.
Dedupe the opening: "You are the orchestrator. Run each step, verify its gate, and only then continue" (front-door paragraph) repeats almost verbatim one paragraph later ("You are the orchestrator... Run steps in order and pass each gate before continuing") — keep one statement and fold the `videos/<project>/` working-directory rule into it.
Tighten the longest chained sentences (e.g., the Step 0 opening rule and the Step 3.1 voice-selection paragraph) into short bullet branches; they each pack 3–4 operational rules into one sentence, which costs re-reading.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense operational prose with almost no explanation of concepts Claude already knows — every step carries commands, gates, or routing rules rather than background. It is not a 5 because of verifiable duplication ("You are the orchestrator. Run each step... pass each gate before continuing" appears in both the front-door paragraph and the following paragraph) and several long, comma-chained sentences that could be tightened without losing meaning. | 4 / 5 |
Actionability | Every step gives copy-paste-ready commands with full flag sets (`npx hyperframes init "videos/<project>" --non-interactive --example=blank --skill=faceless-explainer`, `node <SKILL_DIR>/scripts/audio.mjs sync-durations --audio-meta ./audio_meta.json --storyboard ./STORYBOARD.md`), concrete file formats for `tokens.json`, explicit error policy ("surface stderr and stop — don't pile on recovery commands... cheapest safe edit... then rerun the failed check"), and a documented false-positive with an exact selector. It does not earn below 5: placeholders like `<project>` and `<voice-id>` are appropriately parameterized, and deferring field lists to format references is correct structure, not a gap. | 5 / 5 |
Workflow Clarity | Steps 0–6 are explicitly sequenced with a per-step Goal and a bolded **Gate** (explicit validation checkpoints: "`build-frame.mjs` exited 0 — `frame.md` exists from a named preset", "`lint` and `check` passed and the snapshots were inspected before render"), an error-recovery feedback loop in Step 6, an audio-async dependency handled in Step 5 ("Wait for Step 3.1 audio to finish if audio was started"), and explicit collaborative/autonomous branches at each user-gated step. This matches the top anchor — clear sequence, explicit validation, feedback loops — with nothing missing. | 5 / 5 |
Progressive Disclosure | Structure is good: the body stays an orchestrating overview, all four local `references/*.md` files cited in the body exist, all cited local `scripts/*.mjs` exist, and the Quick Reference table maps each file to when to read it with one-level-deep references. It is not a 5 because `sub-agents/frame-worker.md` is cited twice (Step 5: "this skill's `sub-agents/frame-worker.md`" as half of the worker role payload, and again in the Quick Reference table) but no `sub-agents/` directory exists in the bundle — a broken local reference the Step 5 dispatch depends on. This is a minor organization gap against an otherwise well-signaled structure, so 4 rather than 3. | 4 / 5 |
Total | 18 / 20 Passed |