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.
A high-quality instruction skill: executable commands throughout, explicit validation and feedback loops, appropriate single-reference disclosure, and no padding. The sole deduction is inline version-pinned defaults that will age, though the skill partially mitigates this by deferring to the binary's --help and live model catalog.
Suggestions
Move version-pinned values (default model names, '-m gemini-3.1-flash-lite as of 0.4.5', the dated deep-research agent ID) into references/cli.md, which already states the verified CLI version, and keep SKILL.md pointing at the live catalog ('gemini-api models') instead.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean and assumes competence — no concept explanations, dense command tables, every line actionable — but it embeds time-sensitive version pins inline ("-m gemini-3.1-flash-lite as of 0.4.5", "deep-research-preview-04-2026", dated default model names) rather than confining them to a versioned/deprecated section, which the guidelines penalize. It is noticeably above the mostly-efficient 3 anchor but short of the every-token-earns-its-place 5 anchor. | 4 / 5 |
Actionability | Fully executable, copy-paste-ready commands cover the common cases: the background-agent snippet ("ID=$(gemini-api agent run ... --agent deep-research-preview-04-2026 --background --jq '.id')"), file-path capture ("MASCOT=$(gemini-api image ... --out ./assets/mascot.jpg)"), the complete --jq extraction pipeline, exact flags (--stream=false -o json), and enumerated exit codes with branching guidance. | 5 / 5 |
Workflow Clarity | Sequencing is explicit ("Preflight (once per session, before the first call)"), validation checkpoints are built in (--dry-run "Validate before spending a call", whoami key check), and error-recovery feedback loops are stated ("When a flag in this skill is rejected, trust the binary: run --help on that command and adapt"; branch on exit codes 0/1/2/3). Destructive server-side operations carry explicit user-confirmation requirements, so no cap applies. | 5 / 5 |
Progressive Disclosure | The body is a well-organized overview that clearly signals exactly one one-level-deep reference ("Verified command details, output shapes, and gotchas: `references/cli.md`"). The reference file exists, contains no nested .md references, and the split is appropriate: judgment and gotchas in SKILL.md, command detail in the reference. | 5 / 5 |
Total | 19 / 20 Passed |