Content
63%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 well-organized, executable command reference with genuinely useful agent guidance (dual human/JSON output modes, render-time semantics of draw ops, verification of rendered output rather than project state). Its main weaknesses are duplicated quick-start/examples sections, missing example invocations for most command groups, and a ~250-line monolith that should be split into a lean overview plus a reference file.
Suggestions
Move the per-command group tables into a references/COMMANDS.md (one level deep, clearly signaled) and keep only a quick-start, one full workflow example, and the agent guidance in SKILL.md; this also removes the current duplication between Usage, Examples, and Output Formats.
Add at least one example invocation for the highest-risk command groups (filter add/set with parameters, canvas resize vs. scale, export render presets) so an agent doesn't have to guess flag names or consult the external README.
Fold the validation checkpoints ('verify outputs exist', 'review the rendered image', 'parse stderr on non-zero exit') directly into the workflow examples as numbered steps, creating an explicit render→verify→fix loop instead of a separate tips list.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The command tables are dense, useful reference material with almost no explanation of concepts Claude already knows, but there is noticeable duplication: 'project new -o' appears in Usage, Examples, and again near Output Formats; the --json invocation is shown three times; and the REPL section is explained twice. This fits anchor 3 ('mostly efficient but includes some... padding that could be tightened') better than anchor 4, since several sections could be collapsed. | 3 / 5 |
Actionability | Commands are concrete and copy-paste ready (e.g. 'cli-anything-gimp --project poster.gimp-cli.json draw text --layer 0 --text "Launch Night" --x 96 --y 120 --size 72 --color "#f6f1e8"'), and the Layered Composition Workflow example shows realistic full syntax including negative offsets. It stops short of 5 because most command groups (filter add, canvas resize, media probe) have only one-line table descriptions with no example invocation or flag reference — full syntax is deferred to an external README. | 4 / 5 |
Workflow Clarity | The Layered Composition Workflow gives a clear create→set offsets→draw→export sequence, and the 'For AI Agents' checklist supplies validation checkpoints ('Check return codes', 'Parse stderr for error messages on failure', 'Verify outputs exist after export operations', 'Review rendered outputs after offset, draw, blend-mode, or filter changes'), so the render path is not a blind trust-the-JSON flow. It is 4 rather than 5 because these checkpoints live in a separate tips section instead of being embedded as explicit steps in each workflow, and there is no fix-and-retry loop for a failed render. | 4 / 5 |
Progressive Disclosure | The body has good section structure but inlines roughly 150 lines of command-group reference tables that clearly belong in a separate reference file, and its pointers ('Full documentation: See README.md in the package', 'See TEST.md', 'HARNESS.md in the cli-anything-plugin') reference package files that are not part of this bundle (references/, scripts/, assets/ are absent), so navigation targets cannot be verified. This matches anchor 3 ('content that should be separate is inline; references present but not clearly signaled'). | 3 / 5 |
Total | 14 / 20 Passed |