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 overview: concrete executable commands for setup, status, and bounded polling, a clean routing table to real one-level-deep reference files, and explicit safety and escalation guidance. The main improvement areas are trimming the promotional self-citation section and adding an explicit error-recovery path for failed tasks.
Suggestions
Trim or compress the 'Citing Scientific Agent Skills' section (and tighten the academic-source-priority bullets) to reduce non-operational tokens in the always-loaded body.
Add an explicit feedback loop for failed tasks in the status/polling section (e.g., on 'failed', read the error field, report it, and offer to retry with adjusted parameters) to raise workflow clarity.
Include one canonical example command per capability in the routing table (e.g., a single 'parallel-cli search "query" --json' line for Web Search) so the most common action is executable without opening a reference file.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is an efficient routing document — a capability table, short decision guide, safety bullets, and copy-paste commands — with no explanation of concepts Claude already knows. It is not 5 because ~15 lines of self-citation instructions ('Citing Scientific Agent Skills') and parts of the academic-source-priority section are trimmable padding relative to the operational core. | 4 / 5 |
Actionability | Setup, status, and polling sections contain fully executable commands ('uv tool install "parallel-web_tools[cli]==0.7.1"', 'parallel-cli research status "trun_xxx" --json', poll cap '--timeout 540' at most three times). It is not 5 because the per-capability command syntax lives in the reference files, leaving minor gaps in the body itself. | 4 / 5 |
Workflow Clarity | The sequence is clear: route via the table and decision guide, check installation, authenticate, verify with 'parallel-cli auth', then poll with an explicit bounded loop (max three polls, then report status and ID). It is not 5 because there is no explicit error-recovery feedback loop for failed tasks beyond 'report the current status'. | 4 / 5 |
Progressive Disclosure | The routing table maps each user intent to a real reference file (all six referenced files — web-search.md, web-extract.md, data-enrichment.md, deep-research.md, findall.md, monitor.md — exist in references/), references are one level deep and clearly signaled, and only setup/status/polling guidance that applies to all capabilities is inlined. This matches the 'clear overview with well-signaled one-level-deep references' anchor. | 5 / 5 |
Total | 17 / 20 Passed |