Content
60%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-structured operational methodology: clear phase sequence with named validation checkpoints, concrete output artifacts, and a useful anti-patterns checklist. Its weaknesses are redundant positioning prose, and unspecified execution details — the self-critique questions, scoring mechanics, and the channel/block catalog — that keep phases 5 and 6 from being actionable as written.
Suggestions
Cut the 'How it differs from a quick research router' section and keep only its concrete differentiator list (falsifiable hypotheses, parallel fan-out, per-source files, refresh_targets.md); the slow-vs-rigorous positioning is already made in the intro and the description.
Enumerate the 4 self-critique questions and define the Credibility/Recency/Bias scoring scale (or bundle them in a reference file) so the adversarial pass and triangulation phases are executable without invention.
Add a compact template for sources/NN.md (metadata fields, quote format, score line) so Phase 4's per-source output is unambiguous across runs.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The pipeline, mechanisms, output tree, and anti-patterns are dense and operational, but the positioning ('heavy, methodical end of research', low-risk router contrast) is stated three times — in the intro, the two-paragraph 'How it differs from a quick research router' section, and 'When to use' — which is more than minor trimmable material. Not 2 because nothing explains concepts Claude already knows and there is no filler padding. | 3 / 5 |
Actionability | Concrete rules exist (≥3 differently-typed sources, sources/NN_slug.md naming, Credibility/Recency/Bias dimensions, the output tree), but key execution details are missing: the '4 self-critique questions' are referenced twice without ever being enumerated, the scoring mechanics are unspecified, and the search channels and report blocks live in an external upstream repo rather than the bundle. Not 4 because these gaps exceed 'minor' — the adversarial and scoring phases cannot be executed as written. | 3 / 5 |
Workflow Clarity | The 9-phase table gives a clear sequence with explicit validation checkpoints (Phase 5 triangulation, Phase 6 adversarial pass, Phase 6.5 citation check), a re-evaluation loop in Phase 4, depth gating, and stop-criteria in plan.md. Not 5 because the checkpoints are one-line descriptions without fix/retry mechanics — e.g. 'Lightweight citation check before closing' says nothing about what to check or what happens on failure. | 4 / 5 |
Progressive Disclosure | The body is a well-sectioned overview with one clearly signaled, load-on-demand reference (references/full-catalog.md, verified to exist) — good structure with appropriate placement. Not 5 because that reference's actual content (channels, blocks, phases.yaml) is deferred again to an external GitHub repo, a chained-reference pattern for the sourcing surface. | 4 / 5 |
Total | 14 / 20 Passed |