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 highly actionable operator reference: concrete commands, complete parameter tables with defaults, safety rules for the disabled `add_resource` tool, and a solid troubleshooting matrix. Its main weaknesses are monolithic structure — the config and tool-interface reference belong in separate progressive-disclosure files — and the absence of full validate/retry feedback loops in the setup and troubleshooting flows.
Suggestions
Split the 'Configuration Quick Reference' table and the per-tool 'Tool Interface Reference' into files under references/ (e.g., references/config.md, references/tools.md) and keep SKILL.md as an overview with the tool-selection guide, safety rules, and troubleshooting matrix, linking to the detail files.
Add an explicit validate→fix→retry loop for setup: after `openclaw openviking setup ... --json`, state what a success response looks like and what to re-check (baseUrl, key, /health) before retrying or escalating.
Trim the generic 'Agent Skill best practice' paragraph under `add_skill` — it repeats knowledge Claude already has and adds tokens without skill-specific guidance.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense operational reference material — config tables, tool parameter tables, troubleshooting rows — with essentially no explanation of concepts Claude already knows and no padded prose. Minor trimmable spots exist, e.g., the generic "Agent Skill best practice" paragraph under `add_skill` (frontmatter/tagging advice Claude already knows), so it is efficient but not perfectly lean. | 4 / 5 |
Actionability | Fully executable guidance throughout: copy-paste commands ("openclaw openviking setup --base-url <OPENVIKING_URL> --api-key <API_KEY> --json", the slash-command examples), per-tool parameter tables with required/optional flags and defaults (e.g., `recallLimit` default 6, `openviking_tool_result_read` offset/limit defaults), and symptom→cause→first-action troubleshooting entries. Covers the common operational cases. | 5 / 5 |
Workflow Clarity | Sequencing is explicit ("Read status first" before setup; "branch on JSON error" on failure), and the destructive `memory_forget` flow has a real checkpoint ("If query mode finds multiple candidates, report candidates and ask the user to choose the exact URI; do not delete ambiguous memories"), as does `--force-slot` ("only after user confirms"). It falls short of the score-5 anchor because there is no complete validate→fix→retry feedback loop for setup/health failures — the troubleshooting table points at first actions but not retry-after-fix cycles. | 4 / 5 |
Progressive Disclosure | No bundle files exist (references/, scripts/, assets/ are absent), and the entire ~260-line document is monolithic: the full 20-row config table and the complete tool interface reference are inlined in SKILL.md rather than split into reference files. Section headers are clear and repo docs are listed at the end ("Reference Docs in This Repo"), which lifts it above the 'minimal structure' anchor, but it matches the score-3 pattern of inline reference content that should be separate. | 3 / 5 |
Total | 16 / 20 Passed |