Content
85%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-engineered methodology skill: six gated phases with mandatory artifacts, validation checkpoints, anti-drift and anti-shallow protocols, and a clean one-level-deep reference structure that keeps the deep material out of context. The main cost is token weight — several sections re-explain thinking models, traps, and biases Claude already knows, which pushes ~90KB of references plus a long body against the context budget.
Suggestions
Compress or cut the 'Common Traps', 'Bias Awareness', and 'Complementary Tools Quick Reference' sections to one-line pointers into the existing reference files — the ~60 lines of well-known concepts (sunk cost, 5 Whys, Inversion) are the largest token savings without losing the phase-mapping, which can survive as a single compact table in references/thinking-models-toolkit.md.
Move the Trellis Integration section into its own reference file (e.g. references/trellis-integration.md) and keep only a one-sentence trigger in the body — it is ecosystem-specific dead weight for any non-Trellis project yet occupies ~55 lines of core context.
Tighten the Phase 1 'Essence' and Phase 3 'Ground Truths' key-questions lists to match the template density of Phases 0, 2, and 4 — Phase 1 currently reads as discussion prompts where the other phases specify exact output artifacts.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The core methodology (Phases 0–5 with gates, depth standards, progress tracker) is genuinely prescriptive and earns its tokens, but roughly a quarter of the body explains concepts Claude already knows: the 'Common Traps' section (analogy/complexity/legacy traps), the 'Bias Awareness' table (confirmation, anchoring, sunk cost, status quo, overconfidence), and the 'Complementary Tools' and 'Problem Decomposition' tables (Inversion, 5 Whys, Pre-Mortem, Issue Tree, Fishbone). This is 'mostly efficient but includes some unnecessary explanation' — more than the minor trimming of a 4, but far from the padded explaining-from-scratch of a 2, since each known concept is compressed into a compact table mapped to FP phases. | 3 / 5 |
Actionability | For an instruction-only skill, the guidance is fully executable: copy-paste templates for axioms, the assumption table with exact columns and verdict vocabulary (Keep/Discard/Modify), ❌/✅ example pairs with concrete specifics ('P99 latency must be < 200ms per SLA contract §3.2'), a worked reasoning chain, a complete final-output skeleton, hard minimums (≥3 axioms, ≥5 assumptions, ≥3 ground truths), and literal commands for Trellis integration (`python3 ./.trellis/scripts/task.py add-context ...`). The scoring note says absence of code is not penalized when the guidance is actionable — and here it is, down to the exact markdown to emit for the progress tracker and drift-return message. | 5 / 5 |
Workflow Clarity | The six phases (0–5) are strictly sequenced with a mandatory-artifact gate per phase ('No artifact → no next phase. If a gate is not met, stop and complete it.'), a summary gate table defining what each phase must produce and at what minimum depth, a Phase 5 completion gate of three explicit validation questions (traceability, completeness, honesty) that must all be 'yes', an anti-drift protocol with a running progress checklist and a scripted return-from-tangent message, and anti-shallow depth standards that name what failure looks like. This matches the top anchor: clear sequence, explicit validation, feedback loops, and checklists. | 5 / 5 |
Progressive Disclosure | The body is a well-organized overview with clearly signaled, one-level-deep references: each inline pointer uses a blockquote ('> Deep methodology: `references/axiom-based-reasoning.md`'), all five referenced files exist on disk, no reference file points to further references, and a closing Reference Files table gives content description plus 'When to Read' for each. Long-tail material (12-bias catalog, 15 decomposition frameworks, case studies) is correctly summarized in compact tables and split into the reference files rather than inlined, matching the top anchor. | 5 / 5 |
Total | 18 / 20 Passed |