Content
70%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 an impressively rigorous orchestration document — every phase, gate, and failure path is specified with concrete commands, templates, and validation checkpoints. Its weaknesses are token economy (rhetorical prose and restated external contracts) and structure (a single monolithic file where several detailed contracts belong in references).
Suggestions
Move the restated drydock document-language contract and the detailed C4/Team interaction mechanics into references/ files, keeping only the decision rule and a pointer inline.
Trim rhetorical flourishes ('a new name for a rejected idea is not a new idea', nautical metaphors) to declarative rules — roughly a third of the prose could compress without losing guidance.
Add one compact worked example each of a frontier question round and a ticket with blockedBy edges, to make the abstract protocol terms concrete.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body packs rules densely, but recurring rhetorical flourishes ('signing what was understood, not what was said, is how misreads become load-bearing') and restated external contracts (the full drydock document-language resolution order) are over-explanation that could be tightened or referenced. Mostly efficient, with noticeable trimming opportunities throughout. | 3 / 5 |
Actionability | Concrete anchors are frequent: the executable audit command 'node scripts/shipyard-audit.mjs [repoRoot]' with exit-code semantics, the spec section template, the lesson-to-slot table, exact status enums, and file paths. Minor gaps: no worked example of a frontier question, a ticket, or a completion report to anchor the abstract protocol terms. | 4 / 5 |
Workflow Clarity | Phases 0-5 are explicitly sequenced with validation checkpoints throughout: the blocking yard audit with exit-0/exit-1 consumption, C2/C3 approval stops, the two-axis review gate with repair rule, the three-strike failure halt, and closeout verification plus the sediment checklist. Error-recovery paths (C4 exit and re-entry, session death, fog gate) are explicitly specified. | 5 / 5 |
Progressive Disclosure | Sections are well-organized by phase and cross-skill references are clearly signaled one level deep, but this is a ~180-line monolithic file with no bundle files: detailed contracts (document-language resolution rules, C4/Team mechanics, the sediment slot table) are inlined where reference files belong, and the file itself is long enough to need splitting. | 3 / 5 |
Total | 15 / 20 Passed |