Content
75%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 dense, well-structured body with concrete commands, a clear five-step workflow, and well-organized reference signaling. It scores 4 across the board: lean but with minor redundancy, actionable but referencing scripts that are not bundled, and clearly structured but pointing to reference files that are absent.
Suggestions
Ship the referenced bundles (scripts/sample_size_estimator.py, endpoint_selector.py, phase_gate_scorer.py, onboard.py, ar_evaluator.py; references/*.md; assets/protocol_synopsis_template.md) so the signaled navigation resolves and actionability examples are verifiable.
Add an explicit validate-and-retry checkpoint between tool runs in the Workflow section (e.g., 'if phase_gate_scorer returns REDESIGN, revise the synopsis and re-run steps 2-4') to lift workflow_clarity toward 5.
Trim the redundancy between the intro framing and the Purpose section, and consider moving the inlined forcing-question library to its own reference file, to tighten conciseness and progressive disclosure.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Assumes Claude's competence (no definition of ICH, clinical trials, or statistics) and every section earns its place, but the Purpose section restates the intro framing and the 'seven questions' list restates the onboarding config keys. It is not 5 because those small redundancies could be trimmed; it is not 3 because the body is mostly tight and information-dense. | 4 / 5 |
Actionability | Provides concrete executable commands with real flags ('endpoint_selector.py --input endpoints.json --profile {drug|device|...}', '--sample', '--output {human,json}') and a scripts table. It is not 5 because the referenced scripts/ directory is absent, the examples cannot be verified, and 'sample_size_estimator.py --design {means|proportions|survival} ...' leaves arguments as an ellipsis. | 4 / 5 |
Workflow Clarity | A clear five-step workflow is sequenced with named steps, and the phase_gate_scorer's GO/REDESIGN/NO-GO verdict plus the forcing-question lock-in (1-2 before 3-5) act as checkpoints. It is not 5 because the main tool-run steps lack an explicit validate-then-retry loop, and not 3 because checkpoints are present via the phase-gate gate and the grilling discipline. | 4 / 5 |
Progressive Disclosure | The body is a well-signaled overview pointing one level deep to clearly labeled references (References section lists study_design_canon.md / endpoint_and_power.md / trial_operations.md; Scripts table; Onboarding). It is not 5 because the referenced references/ and scripts/ directories do not exist, so the signaled navigation cannot resolve, and the inlined forcing-question library is long enough to warrant its own reference. | 4 / 5 |
Total | 16 / 20 Passed |