Content
52%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 establishes a clear packaging boundary and a well-sequenced, verifiable 5-step evidence-review workflow with reasonable conciseness. Its main weaknesses are repeated boundary disclaimers that pad the token budget, missing execution details (no examples or report format), and a broken file structure: the four substantive reference files are never linked while the body cites nonexistent external paths.
Suggestions
Link the existing bundle files where they belong in the workflow, e.g. "Classify failures using [failure-taxonomy.md](references/failure-taxonomy.md)" and "Score the output with [review-rubric.md](references/review-rubric.md)".
Remove or correct the dead references (UPSTREAM.md, ../../../scripts/agent-rule-policy.json, ../code-review-excellence/SKILL.md) so navigation matches the actual bundle layout.
Consolidate the four repeated upstream-fidelity disclaimers into a single boundary statement and add a brief output template for step 5's report (scope, confirmed issues, uncertain points, repairs, next observation).
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly terse and assumes Claude's competence, but the upstream-fidelity disclaimer recurs four times ("Do not... claim CrossFrame v5 source fidelity", "State that upstream routing, source continuity and v5 fidelity were not verified", "Do not reuse an upstream grade as proof of correctness") alongside packaging meta-commentary. It could be tightened into a single boundary statement, so it does not reach anchor 4. | 3 / 5 |
Actionability | The 5-step procedure gives concrete directives ("Separate observed facts, interpretations, assumptions and missing evidence", "Provide a specific repair with its source location"), but there are no worked examples, no output format for the final report, and no linkage to the supporting bundle files that would make execution complete. This places it between anchors 3 and 4, short of mostly-executable guidance. | 3 / 5 |
Workflow Clarity | The "Local evidence review" section presents a clearly sequenced 5-step process with explicit verification checkpoints ("Verify any source actually supplied", "Check whether the conclusion follows... which observation would change the conclusion"). It is not 5 because there is no feedback loop or checklist integration, and the existing evidence-boundary-checklist.md and review-rubric.md files that would supply them are never referenced. | 4 / 5 |
Progressive Disclosure | The four bundle files in references/ (case-law.md, evidence-boundary-checklist.md, failure-taxonomy.md, review-rubric.md) are never referenced from the body, leaving ~25KB of directly relevant material undiscoverable, while the body points to paths that do not exist in the bundle (UPSTREAM.md, ../../../scripts/agent-rule-policy.json, ../code-review-excellence/SKILL.md). References are not merely unclear — they are absent, and the dead paths actively misdirect navigation, fitting anchor 2 rather than 3. | 2 / 5 |
Total | 12 / 20 Passed |