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.
A dense, boundary-heavy instruction skill with an unusually disciplined workflow: explicit sequence, runnable validation gates with block conditions, recovery steps, and handoff rules. Its main weakness is bundle alignment — nearly every named reference and script resolves outside the shipped bundle while the local references/ files are orphaned — plus notable rule repetition across sections.
Suggestions
Align the reference graph with the actual bundle: either ship the referenced files (../../references/skills/he-spec/spec-mode-rules.md, spec-artifact-contract.md, the boundary contracts, and the two Plugins/.../scripts/*.py validators) at paths the bundle resolves, or rewrite the body's paths to the existing references/ contents (contract.yaml, evals.yaml, task-profile.json), which are currently never cited.
Consolidate the repeated boundary rules: coding_lens/testing_lens, the runtime persistence boundary, and the strict scope contract are each restated in Procedure, Validation, Stage Arc Boundary, and References — state each once and reference it elsewhere.
Inline one-line trigger conditions for each gate family in step 5 (or point to one reachable reference), since the current "only when triggered" defers to files that are missing from the bundle, leaving the reader unable to act.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean and assumes competence — no concept tutorials, no padding, telegraphic contract-style prose — but rules repeat across sections: coding_lens/testing_lens and the runtime boundary contract each appear in Procedure, Validation, Stage Arc Boundary, and References, and Outputs describes fields in two overlapping paragraphs, so not every token earns its place. | 4 / 5 |
Actionability | It includes concrete validation commands ("python3 Plugins/harness-engineering/scripts/check_bluf_structure.py <spec-path> --json"), enumerated output fields with a list of valid linear_mutation_status values, and two worked input/expected-output examples. Minor gaps remain: gate triggering conditions (step 5's "gate-selection, first-principles, hook, domain, security... gates only when triggered") are never defined inline and the defining references are outside the bundle. | 4 / 5 |
Workflow Clarity | The 9-step procedure is clearly sequenced with explicit validation checkpoints — two runnable lint commands with block conditions, "record each gate as pass, fail, or blocked; never claim readiness from unrun checks", a Failure Mode section with blocked_reason plus one recovery step, and handoff gates before he-plan. This matches the top anchor including feedback loops and error recovery. | 5 / 5 |
Progressive Disclosure | Referenced paths are well signaled (conditional "Read when" clauses, conceptually one level deep), but scored against the actual bundle the structure is misaligned: the named reference files (../../references/skills/he-spec/spec-mode-rules.md, spec-artifact-contract.md, ../../references/*.md, Plugins/.../scripts/*.py) do not exist in the bundle, while the shipped references/ files (contract.yaml, evals.yaml, task-profile.json) are never referenced by name from the body, leaving an unnavitable link graph. It sits above anchor 2 (references are clearly signaled, not buried) but below anchor 4 because most referenced content is not actually reachable. | 3 / 5 |
Total | 16 / 20 Passed |