Content
63%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 content is highly actionable and pedagogically strong — the mental-model section, default-first triage, and security warnings are genuine value — but it is overlong for a SKILL.md. Repetition of the same CLI-limitation guidance and the inlining of feature-level and credential-level detail that belong in reference files cost it both conciseness and progressive-disclosure points.
Suggestions
Deduplicate guidance that appears multiple times: state the MCP-target API-key limitation and the 2LO/3LO CLI-vs-SDK split once (Step 3's auth matrix) and reference it from Path A and Troubleshooting instead of restating it.
Move the semantic-search section, custom-headers section, and the Path D credential/decorator detail into references/ files (e.g., references/search.md, references/credentials.md), leaving the body with the triage table, per-path CLI commands, and one client-code example.
Replace Path D's 'Before you reach for Path D' bullet list with a pointer back to the earlier direct-call justification table — the two lists cover the same five situations.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly AgentCore-specific knowledge Claude would not know, but repeats the same guidance multiple times: the 'API key not supported for MCP targets' limitation appears in Step 3, Path A, and Troubleshooting; the 2LO-vs-3LO CLI limitation is explained nearly verbatim twice; and Path D's 'check if it's actually the right path' bullet list duplicates the direct-call justification table from the earlier 'prefer a Gateway target' section. This matches 'mostly efficient but includes some unnecessary explanation or could be tightened'. | 3 / 5 |
Actionability | Concrete, runnable CLI commands for every path and complete framework-agnostic MCP client code put this above the midpoint, but the Strands snippet references get_gateway_tools/os/asyncio without imports and the LangGraph snippet has undefined llm and an unexplained @app.entrypoint — minor gaps in copy-paste readiness. The 'Mostly executable guidance; concrete code or commands with minor gaps' anchor fits. | 4 / 5 |
Workflow Clarity | Steps 0–3 are clearly sequenced with a triage heuristic routing to Paths A–D, and validation exists (CLI version check, 'agentcore fetch access'/'agentcore status' verification, log-based diagnosis in Troubleshooting). It falls short of the 5 anchor because some checkpoints are implicit — e.g., no explicit 'verify the tool call succeeds after deploy' step, and the 'agentcore logs' recovery guidance is tucked into Troubleshooting rather than the path workflows. | 4 / 5 |
Progressive Disclosure | The one bundle reference (references/policy.md, which exists) is clearly signaled, but the ~550-line body inlines substantial material that belongs in separate reference files — the auth matrix with per-type 3LO/IAM boto3 detail, the semantic-search feature guide, custom-header passing, and the full Path D credential/decorator guide. This matches 'some structure but… content that should be separate is inline'; it is above the buried-reference anchors but below the well-split 4–5 anchors. | 3 / 5 |
Total | 14 / 20 Passed |