Content
63%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 well-structured, actionable readiness-review skill with explicit validation gates and concrete routing. Its weaknesses are cross-section repetition and a References section that points to non-bundle paths while the real bundle file goes unreferenced.
Suggestions
Deduplicate guidance shared across Core Contract, Constraints, Anti-patterns, and Gotchas (e.g., the read-focused rule and the CI-as-readiness warning) so each point lives in one place.
Reconcile the References section with the actual bundle: either reference references/context-preservation.md where context-disposition is invoked, or relink the external Plugins/... paths to verifiable relative bundle paths.
Tighten the three-step Procedure so each step maps to the Validation gates and add a brief fix-and-retry loop instead of only "Stop at the first failed gate."
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly lean and assumes Claude's intelligence, but it repeats the same guidance across sections ("Keep broad reviews read-focused" in Core Contract and Constraints; "Do not treat passing CI alone as merge readiness" in both Anti-patterns and Gotchas), which is more than minor padding; not a 2 because the prose itself is not verbose or explanatory of known concepts. | 3 / 5 |
Actionability | Provides a concrete executable command ("python3 Infrastructure/scripts/validation-and-linting/he_linear_traceability_lint.py <artifact-path>"), explicit routing targets (he-work, autofix, security-ops), and a P0-P3 severity framework; not a 5 because the three-step Procedure remains high-level ("Build the evidence pack from diff, checks, validation, and review threads") without exact mechanics. | 4 / 5 |
Workflow Clarity | A clear sequence (Resolve -> Build evidence -> Rank and emit verdict) is paired with explicit validation gates ("Block go for unresolved P0/P1...", "Stop at the first failed gate", run the traceability lint before a go); not a 5 because there is no fix-and-retry feedback loop and the Procedure steps do not map 1:1 onto the Validation gates. | 4 / 5 |
Progressive Disclosure | Sections are well-organized and a References section signals external files, but per the bundle structure the body references six paths under Plugins/harness-engineering/... that are not present in the bundle, while the one actual bundle file (references/context-preservation.md) is never referenced; not a 4 because the referenced paths are unverifiable and orphaned relative to the real bundle. | 3 / 5 |
Total | 14 / 20 Passed |