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 a dense, highly actionable operator manual: every workflow ships with executable commands, explicit validation checkpoints (publish_result lines, version-conflict handling, purchase consent), and clearly signaled one-level references to live docs and the two bundled scripts. Its main weaknesses are a few over-long inline passages (stale-base protection, client attribution) and limited internal content splitting — most detail lives in SKILL.md itself rather than in bundled reference files.
Suggestions
Split detail-heavy sections (the stale-base protection paragraph, the workspace access modes, and the folder API endpoints) into bundled reference files under references/ and keep one-line summaries with pointers in SKILL.md, matching the one-level-deep pattern already used for the docs URLs.
Break the ~250-word stale-base protection paragraph into a short bulleted workflow (check drift → read live files → reconcile → republish, or --overwrite) so the recovery sequence is scannable instead of buried in prose.
Tighten the client-attribution section to the rule plus one example line, and move the 'Skill version: 1.33.0' line (time-sensitive) out of the body or into a changelog/old-patterns note to reduce maintenance drift.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Nearly all content is service-specific fact Claude cannot know (endpoints, flags, state files, access modes), with no padding on general concepts — e.g. the credentials write command and the three-step publish flow are stated tersely. Not 5: the ~250-word single-paragraph stale-base protection section, the client-attribution persona explanation ("If you are a bot named 'research-bot' running inside Grok Bot..."), and the time-sensitive inline "Skill version: 1.33.0" are instances of over-length that could be trimmed or sectioned. Not 3: these are minor relative to the body's overall density of useful, non-redundant instruction. | 4 / 5 |
Actionability | Copy-paste-ready commands throughout: `./scripts/publish.sh {file-or-dir}`, complete curl requests with JSON bodies for the request-code/verify-code flow, `mkdir -p ~/.herenow && echo "{API_KEY}" > ~/.herenow/credentials && chmod 600`, drive.sh usage examples, and a full publish.sh flags table. Even the no-scripts fallback documents the exact raw API sequence (POST /api/v1/publish with files array, PUT bytes, POST finalizeUrl). Fully executable guidance covering the common cases. | 5 / 5 |
Workflow Clarity | The API-key flow is a 5-step numbered sequence with an explicit save-the-key step; publishing states "A site is not live until finalize succeeds"; `publish_result.*` stderr lines serve as explicit post-run checkpoints (auth_mode, warning, account_url, folder); and version_conflict recovery gives two concrete remediation paths (read live files and reconcile, or --overwrite). Irreversible operations are guarded — domain purchase requires stating price and getting explicit user consent, and --overwrite is gated by stale-base version checking. Not 4: validation checkpoints and error-recovery feedback loops are both present and specific. | 5 / 5 |
Progressive Disclosure | Verified against the actual bundle: both referenced paths (./scripts/publish.sh, ./scripts/drive.sh) exist as real executable here.now scripts, and external references (https://here.now/docs section anchors, openapi.json) are one level deep and clearly signaled with → arrows and a dedicated "read the current docs" trigger list. Not 5: the SKILL.md itself carries ~300 lines of inline detail (full flags table, stale-base mechanics, workspace access modes, folder API details) that could be split into bundled reference files rather than kept inline with offloading only to external URLs; anchor 4 ('most content is appropriately placed; minor organization gaps') fits better than anchor 5's 'content appropriately split'. | 4 / 5 |
Total | 18 / 20 Passed |