Content
78%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 well-engineered body: near-fully-executable code for the core shapes, a symptom-driven workflow with a final validation pass, and an exemplary split of deep material into two well-signaled companion references. The main costs are repetition of the toBeDefined guidance, long quoted passages from Chrome docs, and one test example that depends on an undefined helper. Payload/serialization constraints and version-gated behavior are handled with unusual rigor.
Suggestions
State the 'assert the full response object, not toBeDefined()' guidance once in Authoring and once in the Anti-patterns table, cutting the repetitions in 'How to use' step 4 and the prose after the test snippet.
Define or inline the sendFromContentScript helper (or replace it with a direct chrome.runtime.sendMessage call) so the flagship test example is copy-paste runnable.
Turn 'How to use' step 6 into an explicit feedback loop, e.g. 'if the draft matches any anti-pattern, fix it and re-run the check before committing', to make the validation cycle recoverable rather than a single final pass.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense with non-obvious, Chrome-specific behavior ("a literal `true` (not just a truthy value)", the JSON-vs-structured-clone serialization table, first-listener-wins, the 30-second idle window) and does not re-explain broadly known concepts. It is not a 5 because the "assert the full object, not toBeDefined()" point is repeated four times (step 4, code comment, following prose, anti-patterns table) and long verbatim doc quotes pad sections. | 4 / 5 |
Actionability | Concrete executable sender/listener code for both one-shot shapes, real matcher assertions ("expect(response).toEqual({ statusCode: 200 })"), test-target lists, and an anti-patterns table with concrete fixes. It misses anchor 5 because the test snippet calls an undefined helper "sendFromContentScript({ greeting: 'hello' })", so the primary test example is not copy-paste runnable without it. | 4 / 5 |
Workflow Clarity | "How to use" provides a clear 6-step sequence (identify shape, confirm running, write pair, assert full response, apply guardrails, re-check) ending in an explicit validation step ("Re-check the draft against the Anti-patterns table before committing"), and the worked example acts as a checklist. It falls short of anchor 5 because there is no explicit error-recovery loop spelling out what to do when the check fails, though it clearly exceeds anchor 3's implicit checkpoints. | 4 / 5 |
Progressive Disclosure | The two one-shot shapes stay inline in SKILL.md while the deeper topics are split into two real, verified reference files (references/long-lived-ports.md, references/cross-boundary-messaging.md), each introduced with a description of its contents, listed in References, and one level deep (they link only to external Chrome docs, not further bundle files). This matches anchor 5's clear overview with well-signaled one-level-deep references. | 5 / 5 |
Total | 17 / 20 Passed |