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.
The body is a lean, well-organized rule set for dispatching LinkedIn MCP tools safely: exact tool names per task category, explicit authorization and confirmation checkpoints for every write action, and bounded retry behavior for auth and send failures. Its only shortfalls are the absence of example tool invocations and mild repetition in the write-authorization rules.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Quotes: 'Start with the smallest read operation that can answer the request', 'Distinguish retrieved facts from inference', and the tool lists — terse, operational rules with zero explanation of concepts Claude already knows. The write-authorization pattern is restated across three bullets ('Use it only when the user explicitly asks...') and could be consolidated, which matches anchor 4 ('minor instances ... could be trimmed') rather than anchor 5's every-token-earns-its-place. | 4 / 5 |
Actionability | Quotes: 'Profiles: `get_my_profile`, `get_person_profile`, `search_people`, and `get_sidebar_profiles`' — exact tool names mapped to every task category, plus executable conditions like 'Confirm the final message or connection note unless the user has already supplied it' and 'retry once after it finishes'. No example invocations or arguments are given and 'Keep searches and result pages modest' states no concrete limit, so it falls short of anchor 5's fully executable, example-covered guidance while clearly exceeding anchor 3's pseudocode-level incompleteness. | 4 / 5 |
Workflow Clarity | Checkpoints are explicit for the risky operations: confirm the exact message or connection note before writes, 'dismiss_job ... cannot be undone' is gated on explicit user request, auth has a bounded feedback loop ('report that exact state and retry once after it finishes. Do not loop'), and 'Do not retry a failed write action unless the result proves it was not sent' governs failure recovery. This satisfies the destructive-operation validation requirement, but the overall order of operations (verify request → check session → read → authorized write) is implied across sections rather than sequenced, matching anchor 4 with minor gaps rather than anchor 5's explicit stepwise workflow. | 4 / 5 |
Progressive Disclosure | A self-contained, single-purpose skill (~45 content lines) with three clear sections ('Operating rules', 'Browser and authentication', 'Tool selection'), no nested or buried references — the body references no files and no references/scripts/assets bundle exists — and no content that belongs in a separate file. Per the simple-skill scoring note, well-organized sections alone warrant 5 here; anchor 4 would require organization gaps that are not present. | 5 / 5 |
Total | 17 / 20 Passed |