Content
78%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-structured, highly actionable skill body: copy-paste CLI commands, a sequenced workflow with per-step outputs, and verified one-level-deep references. The main costs are redundancy — the finance-owner guardrail and sibling-skill disambiguation are repeated far more than needed — and a small input-format gap between the markdown template and the JSON inputs the scripts expect.
Suggestions
State the guardrail once (e.g., in Purpose or Anti-patterns) and cut the repeats in the intro, When-to-use, Workflow, and Assumptions sections; also drop the 'Distinct from' table since the frontmatter description already carries that disambiguation — this would tighten conciseness toward a 5.
Bridge the template-to-input gap: either show a minimal program.json / ledger.json / costs.json example inline or state that the markdown template maps to the JSON keys, so the user can go from the assets template to a runnable --input without inspecting the scripts.
Add one explicit validation loop to the Workflow, e.g. 'if burn_runway_tracker flags below-threshold runway or accelerating burn, re-run with an adjusted ramp or escalate before assembling the review packet' — this turns the read-the-output steps into true checkpoints and lifts workflow clarity toward a 5.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and operational — no concept tutorials, every section carries commands or decision rules. Not 5: the 'routes to a named finance owner / never books an entry' guardrail is repeated at least six times across the intro, Purpose, When-to-use, Workflow, Assumptions, and Anti-patterns sections, and the 'Distinct from' table duplicates frontmatter content; not 3: the excess is repeated emphasis rather than padded explanation of things Claude already knows. | 4 / 5 |
Actionability | Concrete copy-paste commands throughout ('program_budget_planner.py --input program.json --profile university-lab --fa-rate 0.585', '--sample', '--output json', the onboarding and autoresearch invocations) plus a Quick examples block. Not 5: a minor gap — the workflow says to fill 'assets/rd_program_budget_template.md' (markdown) while the planner takes '--input program.json', and the input-file schemas are not shown in the body, only reachable via the scripts' --sample/--help. | 4 / 5 |
Workflow Clarity | A clear 5-step numbered workflow with per-step commands and expected outputs ('Read direct / F&A / fully-loaded rollups + assumptions'), an explicit onboarding pre-step, and depth-first sequencing rules for the forcing questions ('Lock 1-2 before opening 3-5'). Not 5: checkpoints are mostly read-the-output rather than explicit validate-fix-retry loops (e.g., what to do when the runway flags fire or the F&A warning triggers is left to the references). | 4 / 5 |
Progressive Disclosure | Clean overview body with one-level-deep, clearly signaled references: each of the three reference files is listed with an annotated scope summary, the scripts table maps tool to purpose/profiles, and every referenced path (references/*.md, scripts/*.py, assets/rd_program_budget_template.md) exists in the bundle. Navigation is easy and nothing is nested two levels deep. | 5 / 5 |
Total | 17 / 20 Passed |