Content
85%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 well-engineered, domain-expert body: every line carries grant-specific guidance Claude could not infer, the multi-step workflow is sequenced with validation checkpoints and feedback loops throughout, and detail is properly pushed one level deep into real, well-signaled reference files. The only deductions are minor: slight redundancy between sections and a few script invocations given without argument syntax.
Suggestions
Show the full command-line signature for `docx_audit.py`, `page_fit.py`, `figure_audit.py`, and `cite_check.py` (as done for `snapshot.py` and `budget_check.py`) so every invocation is copy-paste ready.
Trim overlap between "Common failures" and the corresponding step sections by keeping failures as short pointers (e.g., "stale summary tables — see section 5") instead of restating the detail.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense with non-obvious domain specifics (iCloud "Resource deadlock avoided" placeholders, `~$` lock files, `mv -n`, stale quoted-number labels) and wastes no tokens explaining things Claude already knows. Minor redundancy keeps it below anchor 5: "Common failures" restates several earlier sections and the guiding principles duplicate section 4 content. | 4 / 5 |
Actionability | Provides exact commands with flags (`scripts/snapshot.py <shared folder> <snapshot folder> --tz <PI's time zone> --note "<why>"`, `scripts/budget_check.py JUSTIFICATION.docx --xlsx BUDGET.xlsx`) plus concrete naming conventions (`NN_<document>_vX[_tracked].ext`) and placeholder conventions. Falls just short of anchor 5 because several scripts (`page_fit.py`, `figure_audit.py`, `docx_audit.py`) are invoked without their argument syntax. | 4 / 5 |
Workflow Clarity | Sections 0–9 sequence the full life cycle with explicit validation checkpoints and feedback loops: snapshot and reload before every edit, "resolve each FAIL and each unexplained CHECK" after `budget_check.py`, mock-panel packets confirmed "with checksums", and "verify every new finding against the current files before reporting it". Risky in-place/batch edits are guarded with lock-file checks, snapshots, and never-overwrite rules, matching the anchor-5 pattern. | 5 / 5 |
Progressive Disclosure | Seven reference files (all verified present in the bundle) are clearly signaled in a "read when the step comes up" table and pointed to at the relevant steps; scripts are tabulated with detailed purposes. The core workflow stays inline while templates, prompts, and checklists are appropriately split one level deep — matching the clear-overview anchor exactly. | 5 / 5 |
Total | 18 / 20 Passed |