Content
85%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.
Highly actionable and operationally rigorous: every command is executable, exit codes are turned into branching logic, and risky actions are gated behind approvals with explicit error-recovery loops. The weaknesses are token efficiency — some redundancy across sections — and progressive disclosure: a 388-line monolith whose setup and channel troubleshooting could be split into one-level-deep reference files.
Suggestions
Move the setup, channel-connect, keyless-demo, and paused-state troubleshooting (roughly the first 130 lines) into a one-level-deep reference file (e.g. references/setup.md), keeping SKILL.md to the core inbox workflow and a pointer to it.
Merge the two adjacent keyless-demo paragraphs into one, and consolidate the `--to`/contact-invite semantics that currently appear in the setup, contacts, and useful-flags sections into a single place.
Replace the three verbatim quoted error-message blocks with their exit codes plus a one-line recognition cue, since the exit-code table already establishes 'branch on these, don't parse prose'.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body assumes Claude's competence throughout — no filler explaining what Telegram, Slack, or inboxes are — and nearly every line is operational (exit codes, flags, edge cases). But there are minor trims available: the keyless-demo setup is covered in two adjacent paragraphs ("A no-account (keyless) setup gets an email inbox automatically..." then "The no-account (keyless) setup is a free demo..."), `--to`/contact semantics are restated across three sections (setup, contacts, useful flags), and three full error messages are quoted verbatim despite the later "branch on these, don't parse prose" exit-code guidance. That places it at the efficient-with-minor-overexplanation anchor, not the every-token-earns-its-place level 5. | 4 / 5 |
Actionability | Commands are copy-paste ready and cover the common cases end to end: send/reply/ask/approve/choose invocations, the unattended loop (list → show → reply → resolve) with exact flags, a working exit-code branching shell snippet, and documented `--json` shapes for structured parsing. This matches the fully-executable/copy-paste-ready anchor, above level 4 which permits gaps in coverage. | 5 / 5 |
Workflow Clarity | The unattended loop is given as an explicit sequenced recipe, with validation checkpoints throughout: check unread before writing to a contact, read a conversation when the `unreadBefore` warning fires, branch on exit codes (10 denied / 11 still pending / 1 error), and "check contacts before coordinating between people". Error-recovery feedback loops are present (re-running `channel add` resumes at the missing step, retry delivery once the contact connects), and the "Habits that keep the inbox honest" section functions as a checklist. This matches the explicit-validation/feedback-loop/checklist anchor rather than level 4's minor-validation-gaps. | 5 / 5 |
Progressive Disclosure | The 388-line body is a single self-contained file with no reference files at all, and roughly a third of it (the setup, channel-connect, keyless-demo, and paused-state troubleshooting) is detailed operator-facing material that could live one level deep in a reference file, keeping SKILL.md as a leaner overview. Structure is genuinely good — clear headers, easy navigation — so it sits above level 2's headerless wall of text, but the all-inline placement of content that would benefit from being separate fits the level-3 anchor better than level 4, which expects content appropriately split across files. | 3 / 5 |
Total | 17 / 20 Passed |