Content
71%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 tightly written, highly actionable doctrine: decision tables with first-match-wins semantics, an ordered extension ladder, and concrete promotion and test rules, all without textbook padding. Its main gap is progressive disclosure — the entire doctrine lives in one SKILL.md with no reference files, so longer treatments (D5/naming corollaries, the deciding-table walkthrough) are inlined rather than split out.
Suggestions
Split the deepest sections into one-level-deep reference files (e.g., references/deciding-table.md with worked examples, references/disciplines.md for D1-D5) and keep SKILL.md as the overview plus the short version, surfacing each with a clearly signaled 'See X' link.
Add an explicit feedback checkpoint to the decision workflows, e.g. 'after walking the table, re-check the chosen rung against the violated anti-goal and re-walk if the anti-goal was threatened, not crossed' — a validate/retry loop for the deciding process.
Trim rhetorical framing lines ("The asymmetry is brutal — design from it", "That's the point") and the closing eight-bullet restatement, or compress 'The short version' into a single pointer to the section headers.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense original doctrine with no padding over concepts Claude already knows — every section states a rule and its consequence. Minor trimmable overhead remains (rhetorical flourishes like "The asymmetry is brutal — design from it" and the closing 'short version' restating all eight rules), which keeps it at 'efficient; minor instances of over-explanation' rather than lean-and-perfect. | 4 / 5 |
Actionability | The core procedures are concrete and executable: a first-match-wins deciding table with explicit outcomes, an ordered three-rung extension ladder, the >=2-internal-consumers promotion gate, and a named test-discipline rule. Per the rubric's instruction-only note the absence of code is not penalized; a few directives remain abstract ("Invest heavily before a name escapes its file") and D5 largely defers to $naming, so it stops short of fully-executable guidance. | 4 / 5 |
Workflow Clarity | Multi-step decision processes are clearly sequenced with explicit ordering semantics ("walk these in order. First match wins"; "Reach down only when the rung above doesn't fit") and consequences for each branch (promoting too early vs. too late). No validate-and-retry feedback loop is needed for a design-doctrine skill with no destructive operations, but no explicit re-check loop exists either, placing it at 'clear sequence with most checkpoints present'. | 4 / 5 |
Progressive Disclosure | There are no bundle files at all (references/, scripts/, assets/ are absent), so all ~265 lines of doctrine — five full disciplines, the deciding table, the promotion contract, and test doctrine — live inline in SKILL.md. Cross-skill references ($sdk-seam, $naming, $pedantic, $etiology) are clearly signaled and section headers are good, but content that naturally belongs in separate reference files (e.g., D5, which already says 'See $naming for the full treatment') is inlined, matching 'some structure but... content that should be separate is inline'. | 3 / 5 |
Total | 15 / 20 Passed |