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.
The body is a highly actionable, command-first guide with genuine error-recovery detail around connection states. Its main weaknesses are an incoherent step-numbering scheme ('skip to Step 2' with no Step 2), no validation guidance for destructive actions like delete-store, and a single-file layout that inlines a reference-grade action catalog instead of splitting it out.
Suggestions
Fix the broken workflow numbering: either label sections Step 1/Step 2 explicitly or replace 'skip to **Step 2**' with a real heading reference.
Add a validation checkpoint before destructive actions (e.g., confirm the store ID with get-store and echo-back confirmation before delete-store, since deletion is irreversible).
Move the 19-row action catalog into a references/ file (or trim to the top few) since 'membrane action list --intent' already provides live discovery, keeping SKILL.md as the overview.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dominated by executable commands, tables, and state-machine details that earn their tokens, with only minor over-explanation that could be trimmed: the intro paragraph explaining what BTCPay Server is ("a self-hosted, open-source cryptocurrency payment processor"), the filler line "Use action names and parameters as needed", and a garbled agent-types note. This matches anchor 4 ('Efficient; minor instances of over-explanation'); it is not a 3 because the padding is a handful of lines out of ~165, and not a 5 because those unnecessary passages exist. | 4 / 5 |
Actionability | Every workflow stage has copy-paste-ready commands (npm install, membrane login, connection ensure with --wait polling, action list/run, proxy with a full flag table). Minor gap: the run example uses a generic --input '{"key": "value"}' placeholder with no real BTCPay example (e.g., create-invoice with an amount and storeId), and one command inconsistently switches from membrane to npx. This matches anchor 4 ('Mostly executable guidance... with minor gaps'); not a 5 because the common-case invocation is shown only schematically. | 4 / 5 |
Workflow Clarity | The sequence (install → login → connect → poll for READY/CLIENT_ACTION_ERROR states → search → run) is present and the connection polling includes error handling, but the text tells the reader to 'skip to **Step 2**' when no Step 2 heading exists (numbering is incoherent: only '1b' appears), and destructive operations such as delete-store are listed with no verification guidance. Per the rubric cap for destructive/batch operations lacking validation, workflow clarity cannot exceed 3. | 3 / 5 |
Progressive Disclosure | The file is cleanly sectioned and easy to navigate, but it is a ~165-line monolith with no bundle files at all: a 19-row action catalog and a proxy flag reference that duplicate what 'membrane action list' and CLI help provide are inlined rather than split into a references file. This matches anchor 3 ('content that should be separate is inline'); it is not a 4 because reference-type material is not placed in separate files, and not a 2 because the structure and section headers are clear. | 3 / 5 |
Total | 14 / 20 Passed |