Content
82%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, disciplined overview of a bespoke domain: no filler, hard limits and exact tool tags throughout, and an explicit tool order with embedded checks and error-refusal guidance. The residual gaps are deferred tool-call syntax, absent lint-failure recovery steps, and the unlinked references/README.md.
Suggestions
Include one complete worked tool call (e.g. a full fabula_derive or thym_run invocation with all its tags filled in) so guidance is copy-paste ready instead of deferring syntax to "the Tools section".
State what to do when cosm_lint reports failures — fix in place, report to the user, or re-run after edits — to close the validation loop for the reading workflow.
Link references/README.md from the "More" section so the bundle's source map is discoverable from the body rather than partially duplicated.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The ~175-line body is dense with novel domain rules and contains no padding and no explanation of concepts Claude already knows ("People are never in files. Employees live only in Comprise's sealed database."); compatibility with old versions ("oes: emf/1… is still read", "a program that still says `tell`… runs the same, but write `entail`") is handled inline as explicit deprecated-spelling notes, so every token earns its place. | 5 / 5 |
Actionability | Guidance is concrete — exact tags and values ("`thym_run`, only when asked: `<program>`, `<path>` the first input, others as `<inputs>` lines (`name=path`)", "`fabulac` `<do>build</do>`", "`<do>compile</do>` with `<what>` `code`… or a catem's depth: `line`, `overview`, `manual`") plus hard limits (6,000/24,000 characters, 60 files, 6,000 lines) — but exact tool-call syntax is deferred ("Their tags are in the Tools section") and no complete worked example is shown, so it is mostly executable with minor gaps rather than copy-paste ready. | 4 / 5 |
Workflow Clarity | "The tools, in order" gives a numbered sequence with explicit checkpoints (`cosm_lint` at step 2, "`thym_check`… Use it first" before `thym_run`, fabulac job "`status` says what landed; `cancel` stops it; `build` again resumes it") and refusal handling ("say what was refused and what would make it run"); however, what to do when `cosm_lint` reports failures is never addressed, leaving a minor validation gap that keeps it below the 5 anchor. | 4 / 5 |
Progressive Disclosure | The body is a genuine overview with clear sections pointing one level deep to the specs ("`docs/MACROCOSM_OES.md`… `docs/THYM.md`, `docs/ASCRIBE.md`, `docs/COMPRISE_OES.md`"), and the bundle's `references/README.md` is a well-organized source map; but the body never links `references/README.md` directly (the "More" section duplicates part of it), a minor organization gap versus the 5 anchor's fully well-signaled references. | 4 / 5 |
Total | 17 / 20 Passed |