Content
68%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 lean, actionable two-step workflow with an executable command, an input specification, and a well-signaled reference to the report template. Its main weakness is the absence of validation/error-recovery for a batch download operation, which the rubric caps at 3 for workflow clarity.
Suggestions
Add a validation checkpoint after artifact collection: verify `comparison-artifacts/` is non-empty and contains the expected branch/stimulus/model subdirectories before proceeding to insight extraction.
Add an error-recovery branch (e.g. what to do if the collect script fails or the input JSON path is missing), since this is a batch operation that fails silently without verification.
Show a concrete example of how to read a trajectory markdown file and turn it into an answer, so the 'Extract insights' step is fully executable rather than implied.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly efficient: a tight two-step procedure with a concrete command and a structural diagram, with only minor over-explanation (e.g. spelling out what a trajectory file records). | 4 / 5 |
Actionability | It provides an executable command (`npm run compare:collect -- input.json`), the expected input shape, output directory, and a folder-structure diagram, with only minor gaps (no example of how to actually read/parse a trajectory file). | 4 / 5 |
Workflow Clarity | Two steps are clearly sequenced and there is a conditional check (skip collect if the directory exists), but this is a batch download operation with no validation that artifacts were retrieved successfully and no error-recovery checkpoint, which caps the score at 3 per the rubric. | 3 / 5 |
Progressive Disclosure | Structure is clean: the body is an overview with the report template correctly delegated to a one-level-deep, clearly signaled reference (`./references/report-template.md`); the only minor gap is that the referenced template is quite thin. | 4 / 5 |
Total | 15 / 20 Passed |