Content
65%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 knowledge-dense, largely actionable framework with strong heuristics and executable formulas, weakened by the absence of an explicit step-by-step workflow and by data-access and data-shape gaps that leave the more advanced functions un-runnable as written. The monolithic inline structure also leaves no room for progressive disclosure of the more encyclopedic tables.
Suggestions
Define the expected input shapes (or a data-fetch step) for 'yield_sustainability' and 'risk_adjusted_yield' so every code block is executable as written.
Add an explicit numbered workflow (gather rates from DeFi Llama → assess sustainability → risk-adjust → render output) with a sanity-check step on quoted APYs before recommending.
Move the time-sensitive protocol/TVL tables and data-source details into a reference file (e.g., references/protocols.md) and keep SKILL.md as the framework overview.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense reference material — tables of thresholds, executable formulas, and evaluation criteria — with almost no filler; only minor over-explanation of broadly known concepts (e.g., restating the impermanent-loss formula with a worked example) keeps it from the 'every token earns its place' anchor at 5. | 4 / 5 |
Actionability | Mostly executable: 'impermanent_loss', 'net_lp_yield', and 'lending_rate_signal' are complete, copy-paste-ready functions with concrete thresholds. It falls short of 5 because 'yield_sustainability(protocol)' and 'risk_adjusted_yield(opportunities)' consume object shapes (e.g., 'protocol.annual_fee_revenue', 'opp.real_yield_pct') that are never defined, and the data-source table gives no commands or API calls to actually obtain the inputs. | 4 / 5 |
Workflow Clarity | The analysis flow is implied (survey yield sources → fetch data → apply risk adjustment → render the output template) but never explicitly sequenced; no numbered steps or checkpoints exist between the concept sections and the output format. It is not a 2 because the output template and per-module heuristics do define what a complete analysis looks like. | 3 / 5 |
Progressive Disclosure | No bundle files exist and the body is a single ~260-line inline document. It is well-sectioned with headers and tables, but time-sensitive content (protocol TVL ranges, staking-ecosystem tables, data-source details) that belongs in a separate reference file is inlined, and the skill is far over the 50-line threshold that would let a self-contained skill score 5 on organization alone. | 3 / 5 |
Total | 14 / 20 Passed |