Content
77%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 disciplined index: routing tables, a sequenced 11-phase loop with explicit validation gates and an error-recovery escalation path, and verified one-level-deep references carrying all the detail. Its main weakness is rhetorical repetition — the 'read the references' directive is restated in five-plus variations, including a full exhortation section and a closing line that repeats earlier ones.
Suggestions
Collapse the '🚨 READ THE REFERENCES. THIS IS NOT OPTIONAL' blockquote, the gate rule, and the closing 'The references are the skill' line into a single two-sentence statement — the directive currently appears in five-plus variations across the file.
Trim the 'What to Do Right Now' section to only the steps not already covered by the routing tables (identify runtime → open its reference, then Phase 0), or cut it entirely since the phase table already sequences entry.
Reduce the Specialist Tools framing ('Failing to use these tools in their domain is a process failure, not a stylistic choice' plus the two 'you are doing it wrong' examples) to one sentence.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The index structure is mostly efficient, but the 'read the references' directive is repeated in five-plus forms — 'READ THE REFERENCES. THIS IS NOT OPTIONAL', 'Every reference below is mandatory when its scenario applies', 'The gate rule', 'Failing to use these tools in their domain is a process failure', and the closing 'The references are the skill. This file is an index.' which restates the earlier 'The knowledge is in references/'. The 'What to Do Right Now' section also restates the routing tables above it. | 3 / 5 |
Actionability | Routing tables give exact files to open per runtime, tool, phase, and situation, plus one copy-paste discriminator command ('du -h ./target' and "strings -n 12 ./target | rg -i 'bun|node_modules|...'") and a numbered 'What to Do Right Now' start sequence. Below 5 only because nearly all executable commands live in the references — the body itself contains just the one discriminator. | 4 / 5 |
Workflow Clarity | Phases 0-10 are clearly sequenced with one reference each and explicit validation gates: Phase 6 'confirmed only when toggling the suspected cause toggles the bug', Phase 7 'Fix locked by a failing proof', Phase 10 'four evidence gates before declaring done', and cleanup 'verify git diff shows only fix + test'. A feedback loop is present (Oracle Triple 'after 2 consecutive failed rounds') alongside the safety-invariant checklist. | 5 / 5 |
Progressive Disclosure | The body is a pure index with well-signaled one-level-deep references: all 24 referenced paths (runtimes, tools, methodology, scripts) verified to exist in the bundle, ~4,300 lines of detail appropriately split across 21 reference files and 3 scripts, and routing tables by runtime/tool/phase make navigation trivial. | 5 / 5 |
Total | 17 / 20 Passed |