Content
27%Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
This skill is overly verbose, repeating nearly identical fetch-and-parse patterns for five different sources without abstracting them into a reusable template. While it provides concrete URLs and tool commands, the actionability is undermined by speculative CSS selectors and unclear tool availability. The monolithic structure with significant redundancy makes it a poor use of context window tokens.
Suggestions
Abstract the repetitive per-source fetch pattern into a single template (e.g., 'For each source: 1. Check actionbook, 2. Try agent-browser with selector, 3. Fall back to WebFetch') and list sources as a compact table with URL, selector, and result format columns.
Extract the detailed per-source instructions and error handling table into a separate reference file (e.g., SOURCES.md) to reduce the main skill to an overview with clear navigation.
Add explicit validation feedback loops: after each fetch, check if results are non-empty and within the time range before proceeding, with concrete conditional logic rather than vague 'retry with alternative tool' instructions.
Remove or significantly compress the Tool Chain Priority section since it restates what's already shown in the per-source instructions — a single sentence stating the priority order would suffice.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The skill is extremely verbose at ~180 lines. It repeats similar agent-browser/WebFetch patterns for each source with minimal variation, includes redundant tool chain priority tables that restate what's already shown inline, and explains execution mode detection logic that could be much more compact. Much of this content could be collapsed into a single template pattern. | 1 / 3 |
Actionability | The skill provides concrete commands and URLs, but the code blocks are pseudocode-like rather than truly executable — `agent-browser` CLI commands and `mcp__actionbook__search_actions()` calls are presented without clear context on whether these tools actually exist or how they're invoked. CSS selectors like `.Post` and `article` are speculative and may not work on the actual sites. | 2 / 3 |
Workflow Clarity | The two execution modes (agent vs inline) are clearly distinguished, and there's a logical sequence for fetching sources. However, the validation section is minimal — it says 'retry with alternative tool' and 'report reason' but lacks explicit feedback loops (e.g., check result → if empty → try fallback → if still empty → mark as failed). Time filtering is mentioned but not explained how to actually implement it. | 2 / 3 |
Progressive Disclosure | All content is monolithically inlined in a single file with no bundle files. The repetitive per-source fetch instructions could easily be extracted into a reference file or template. The tool chain priority section duplicates information already presented in the per-source sections. No external references are provided despite the content length warranting them. | 1 / 3 |
Total | 6 / 12 Passed |