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.
The body is an exceptionally actionable operational contract: exact tool calls, payloads, commands, numeric targets, speech-validation gates, and bounded feedback loops for both modes, with no filler or teaching of known concepts. Cross-section rule duplication and a deployment-path-only reference to the bundled metrics script keep it just short of top marks on conciseness and progressive disclosure.
Suggestions
Deduplicate rules stated twice — the voice-lock rule ("re-read that file before EVERY call") and the no-resubmit rule ("Retry only failed takes; never resubmit completed ones") both reappear in 'Hard rules'; state each once and keep 'Hard rules' as one-line recaps without re-explaining.
Reference the bundled script by its bundle-relative path (scripts/speech_metrics.sh, matching its own usage header) alongside the ${HF_WORKFLOWS} deployment path, so the reference is discoverable from the skill folder itself.
Trim motivational rationales where the rule is already emphatic (e.g. "A remembered-not-reread pair is exactly how a video ends up with different voices per block") to bare directives.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Dense operational directives with no teaching of concepts Claude already knows, and the explanatory asides carry unknown domain facts ("Every period ≈0.7s and comma ≈0.5s of dead air", "The script trims the provider's head/tail padding"). Not 5: several rules are stated twice across sections ("re-read that file before EVERY call" vs "ONE voice everywhere... re-read from voice.lock"; "Never resubmit the whole batch because one line failed" vs "Retry only failed takes; never resubmit completed ones"), and some motivational rationales ("A remembered-not-reread pair is exactly how a video ends up with different voices per block") could be trimmed. Not 3: there is no real padding or over-explanation. | 4 / 5 |
Actionability | Fully executable, copy-paste ready: exact tool calls and payload shapes ("generate_audio_batch. Every item is {index, params:{model:\"seed_audio\", prompt, voice_id, voice_type, format:\"wav\", count:1}}"), an exact prompt template ("[ {DELIVERY}, {optional line mood}, starts speaking immediately] [00:00-00:09] {line}"), an exact sandbox command ("bash ${HF_WORKFLOWS}/.../speech_metrics.sh work/voices/voice01.wav"), and an exact join command ("ffmpeg -f concat -safe 0 -i parts.txt -c copy work/voices/narration.wav"), all with concrete numeric targets (9.4–9.8s window, ~30–34 words, groups of six, timeout_seconds:25, 2048-char limit, ~1800-char chunks). | 5 / 5 |
Workflow Clarity | Clear sequences per mode with explicit validation gates ("`speech=` must land in the window; `pauses=` must be 0"), feedback loops ("Out of window → REWRITE THE TEXT and regenerate. Never atempo..."; "Retry only failed takes; never resubmit completed ones"), error-recovery paths ("If all_terminal:false, wait poll_after_seconds and call it again"), and anti-loop protections ("If the group shows no status change for 20 minutes, return its pending indices/job ids to the caller instead of looping silently"; "Budget ~3 attempts per line... never loop"). This batch workflow has validation and retry handling throughout, so the batch-cap at 3 does not apply. | 5 / 5 |
Progressive Disclosure | Well-sectioned single file (Inputs/outputs, tool contract, Mode A, Mode B, Hard rules, Reporting, Safety) where all inline content is needed at runtime; the bundled script exists (scripts/speech_metrics.sh, real and functional). Not 5: the body never points to the bundle by its own path — the script is referenced only via its deployment path "${HF_WORKFLOWS}/faceless-channel-video/scripts/narrator/speech_metrics.sh", and assets/icon.svg is unused, so references are not well-signaled relative to the actual bundle structure. Not 3: structure is strong and navigation within the file is easy. | 4 / 5 |
Total | 18 / 20 Passed |