Content
92%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 investigation protocol with concrete commands, a mandatory verification gate, and clean one-level-deep reference disclosure. Its only weakness is minor cross-section redundancy that lightly inflates the token budget.
Suggestions
Consolidate the approval-gate rules: §1.10 defines them fully and §8 / §6 only need a short back-reference ("under the §1 approval gate") rather than restating conditions.
Fold the §1.6 empty/404 correlation rule into the §6.3 runtime-evidence item by reference, since both cover the same wrong-key vs. absent distinction, to remove duplicated wording.
The §4 routing branches repeat the grep pattern inline; stating the canonical `grep -rlF` invocation once in a short 'Routing commands' line and referencing it would trim tokens without losing actionability.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Lean, prescriptive protocol that assumes Claude's competence and avoids concept re-explanation, but carries minor redundancy — the approval gate (§1.10) and runtime-evidence rules (§1.6) are restated in §8 and §6.3. | 4 / 5 |
Actionability | Provides concrete executable commands in documented form — `uip docsai ask "<question>" --source docs`, `grep -rlF "<signal>" references/ --include="*.md"`, `or jobs logs <key> --output json`, `maestro <type> instance incidents <id>` — plus specific tooling (TaskCreate/TaskUpdate, AskUserQuestion) and concrete file paths. | 5 / 5 |
Workflow Clarity | Clearly sequenced ten-phase workflow (Anchor → Extract → Route → Walk → Verify → Escalate → Present) with an explicit mandatory six-item verification checklist, retry caps, and feedback loops ("Any check fails → ONE targeted re-fetch", "Still failing → insufficient evidence"). | 5 / 5 |
Progressive Disclosure | Overview body points one level deep to real, phase-signaled references — `references/investigation_guide.md`, `references/summary.md`, `references/escalation.md`, `references/presenting.md`, and per-domain `investigation_guide.md` / `*/playbooks/` files (all verified present) — with detail appropriately split out. | 5 / 5 |
Total | 19 / 20 Passed |