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.
A thorough, well-structured router body with concrete CLI commands and deterministic, ordered classification logic. Its main weaknesses are redundancy between the prose 'By task' section and the 'Pick a track' table, and that several referenced files (RULES.md, peers.yaml, peers.schema.json, sendbird-data-migration.md) are cited but not bundled.
Suggestions
Collapse the 'By task' prose section and the 'Pick a track' table into a single source of routing truth to remove the repeated precedence rules; keep one as a quick-reference table and link to it from the other.
Bundle (or inline the essentials of) the referenced files RULES.md, peers.yaml, peers.schema.json, and sendbird-data-migration.md so the one-level-deep references actually resolve.
Add one short worked routing example end-to-end (input -> matched signal -> chosen track -> invoked sub-skill) to remove residual judgment on edge cases.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Dense and free of beginner-concept padding, but the 'By task' prose section and the 'Pick a track' table restate the same routing decisions, and precedence rules ('SDK/engine signal beats OS-target', 'peer signals take precedence over web rows', 'audit intent beats CLI check') are repeated 3-4 times - noticeable redundancy that could be tightened. Not a 4 because the repetition is more than minor; not a 2 because it never explains things Claude already knows. | 3 / 5 |
Actionability | Concrete executable commands in the CLI table (getstream init/api/env/token/login/skills, getstream -h) and deterministic routing signals with explicit examples ('swift / .xcodeproj -> stream-swift'); an instruction-only router, so absent code is acceptable. Minor gaps: routing still requires some judgment on edge cases and no worked end-to-end example is shown. | 4 / 5 |
Workflow Clarity | Classification is explicitly ordered ('Scan the user input for the signals below in order ... deterministic - no probes, no fetches, no CLI checks at this stage') with stated precedence rules, an explicit disambiguator step, and a clear install-if-missing -> invoke hand-off. Not a destructive/batch skill so the validation cap does not apply; falls short of 5 because precedence is restated across prose and table, which could blur the exact ordering. | 4 / 5 |
Progressive Disclosure | References are one-level-deep and clearly signaled via markdown links ([RULES.md], [peers.yaml], [peers.schema.json], [sendbird-data-migration.md]) and content is conceptually well split, but none of these referenced files are present in the bundle (no references/scripts/assets dirs, only SKILL.md), so the disclosed material is not actually available - the structure is good yet unsupported by the bundle. | 3 / 5 |
Total | 14 / 20 Passed |