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.
The content is a dense, disciplined routing document: two clearly separated modes, condition-driven reference selection, explicit post-write verification with invalidation rules, and a properly externalized one-level-deep rule bundle. Its only real costs are repeated boilerplate phrasing across the reference-selection and verification bullets and a few overly long sentences that demand re-reading.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body explains nothing Claude already knows and assumes competence throughout — every line is a directive such as "Read only the references whose conditions below match the affected files". It falls short of anchor 5 because the boilerplate "read [X] through EOF and add each applicable rule to the checklist" repeats across ~8 bullets and the parallel "explicitly verify every applicable..." bullets (lines 55–63) could be consolidated into a single stated convention, costing tokens without adding information. | 4 / 5 |
Actionability | Guidance is concrete and executable for an instruction-only skill: precise conditions for reading each reference ("Before adding or changing a Java throw statement... read the Java precondition rules"), a defined checklist mechanic, and a concrete evidentiary standard for HighFrequencyInvocation ("Invocation after each DDL or explicit metadata refresh request proves reachability, not high frequency by itself"). One step remains generic — "Run the applicable formatting and standards checks required by the repository workflow" names no commands — a minor gap that keeps it below anchor 5. | 4 / 5 |
Workflow Clarity | Both modes are clearly sequenced: pre-work reference selection, checklist construction, implementation, post-write verification, then repository checks. Validation checkpoints and feedback loops are explicit — "After the last write, manually recheck the effective task delta", "Treat passing Checkstyle and Spotless as evidence only... neither result replaces the manual recheck", and "A later write invalidates final-check evidence only for affected files; recheck those portions before verification or handoff" — matching the anchor-5 pattern of explicit validation with error-recovery guidance. | 5 / 5 |
Progressive Disclosure | The body is an orchestration overview with 13 well-signaled, condition-gated markdown links to references/rules/*.md and references/standalone-audit.md, all of which exist and are exactly one level deep; the rule content itself lives in those files rather than inline. standalone-audit.md further points to scripts/build_audit_inventory.py (which exists, with tests), and each reference is tied to an explicit condition for when to read it, making navigation easy. | 5 / 5 |
Total | 18 / 20 Passed |