Content
92%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 excellent example of lean, high-signal instruction writing: concrete rules, explicit rework loops, and a precise output format, with no filler. The sole material defect is progressive disclosure — it directs the loader to a `disclosure.md` file that is absent from the bundle and referenced without a path.
Suggestions
Add the referenced `disclosure.md` to the bundle (e.g. as references/disclosure.md) or correct the reference to wherever the disclosure rules actually live, since the body routes organic/email/creator content to it.
Use explicit relative paths or links when pointing to bundle and sibling files (e.g. [disclosure.md](references/disclosure.md), ../claims-gate/SKILL.md) so referenced files are clearly signaled and verifiable.
If per-platform rule sections grow further, split them into one-level-deep reference files (e.g. references/meta.md, references/tiktok.md) with clear links, keeping SKILL.md as the overview.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense, telephonic rule statements with zero concept explanation — e.g. "no fake play buttons, close-X, notification badges, fake DMs" and a compact category table — and supplies non-obvious specifics ("melatonin is prescription-only in the UK and Australia"). Every token is a decision rule Claude could not reliably reconstruct, matching the 'lean and efficient' anchor. | 5 / 5 |
Actionability | For an instruction-only skill the guidance is fully concrete: banned vs. approved wordings ("supports restful sleep" vs. "treats", "cures", "for insomnia"), an explicit rewrite strategy (situation framing instead of "you have…"), and a precise report format (`pass (Meta, TikTok)` · `fixed (what, why)` · `flag (risk, rule, decision needed)`). Per the rubric's scoring note, absence of code is not penalized when the guidance is this actionable. | 5 / 5 |
Workflow Clarity | A clear five-step sequence (Category first → Per ad, per platform → Disclosures → Fix → Report) with gating rules ("Gambling, dating, political → most platforms need pre-authorisation: flag and stop"), an explicit feedback loop ("fix that; re-run the claims check and this check"), and verification discipline ("Never 'pass' a rule you could not check — say 'not checked: <rule>, <why>'"). This matches the top anchor's explicit validation and error-recovery loop. | 5 / 5 |
Progressive Disclosure | Sections are well organized, but the body instructs "load `disclosure.md` instead" and no such file exists anywhere in the bundle (no references/, scripts/, or assets/ directories) — a dangling, unpathed reference. A missing referenced file plus bare inline mentions of sibling skills fits 'references present but not clearly signaled' (3) better than the minor-gaps profile of 4. | 3 / 5 |
Total | 18 / 20 Passed |