Content
53%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 admirably lean overview with a logical workflow, but it is a pointer document whose pointers mostly fail: over half the referenced files don't exist, the existing bundle files are unlinked, and with no code or concrete steps the skill provides little standalone actionable guidance.
Suggestions
Fix the reference list to match the actual bundle: replace the broken concepts/, examples/, and troubleshooting/ links with the real files (e.g. connection.md for connection architecture, quickstart.md for lifecycle, and add webhooks.md and environment-variables.md), or restore the missing directories.
Add one concrete anchor of guidance in the body — e.g. a minimal SDK quickstart snippet or the exact webhook→URL→connect sequence — so the skill is actionable even before opening references (this is the skill's biggest actionability gap).
Add an explicit validation checkpoint in the workflow (e.g. 'Verify the webhook event and signature validation arrive before opening the WebSocket; if the connection drops, resume from the heartbeat/reconnect state rather than re-subscribing') to close the workflow_clarity gap.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The 27-line body is lean with zero padding: it never explains what RTMS is, what WebSockets are, or any concept Claude already knows, and the workflow steps are single dense lines ("Implement the WebSocket lifecycle first: event subscription, connection validation, media start, heartbeat, reconnect, and shutdown"). Every token earns its place; it clearly matches the score-5 anchor. | 5 / 5 |
Actionability | No code, commands, URIs, or executable steps appear anywhere in the body — steps like "Process media types intentionally" and "Design downstream pipelines for buffering, transcription, AI analysis" describe rather than instruct. It sits above the score-1 anchor because the steps do enumerate concrete subtopics (lifecycle stages, media types, debug isolation areas), matching "high-level hints but missing the specific steps to execute". | 2 / 5 |
Workflow Clarity | The six steps are sensibly ordered (surface → approach → lifecycle → media → pipelines → debug) and step 6 acts as a soft error-recovery pointer, but there are no validation checkpoints (e.g. verify the webhook/authorization handshake before subscribing) and each step defers all execution detail. This fits "Steps listed but validation gaps; sequence present but checkpoints missing or implicit"; the operations are not destructive, so the cap-at-3 rule is not triggered but score 4 would require explicit checkpoints. | 3 / 5 |
Progressive Disclosure | Scored against the actual bundle: of the 7 paths referenced, concepts/connection-architecture.md, concepts/lifecycle-flow.md, examples/rtms-bot.md, and troubleshooting/common-issues.md do not exist, while four real files in references/ (connection.md, quickstart.md, webhooks.md, environment-variables.md) are not discoverable from SKILL.md at all, and the linked full-guide.md chains onward to the same broken paths. The one-level-deep reference list is well-formatted but the majority of navigation targets are broken and half the bundle is orphaned, fitting "references are buried / minimal structure" rather than the cleanly navigable score-4/5 anchors. | 2 / 5 |
Total | 12 / 20 Passed |