Content
88%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 strong routing-style SKILL.md for a very broad domain: executable commands with real model IDs, explicit validation checkpoints in every spelled-out workflow, and a verified, clearly signaled reference bundle. The main costs are repetition of the maxTokens warning, one hard-to-parse Harness paragraph, and two-level reference nesting in the migration procedures.
Suggestions
Consolidate the maxTokens warning to the Critical Warnings section and reference it with a one-line pointer from the Converse, Examples, and ThrottlingException sections instead of restating the full rationale each time.
Break the long Harness paragraph into a short numbered checklist or table (create-harness inputs → poll get-harness until READY → invoke with runtimeSessionId/messages → read streamed events → agentcore CLI) so each requirement is scannable instead of buried in a run-on sentence.
Flatten the two-level reference nesting for the KB and agent migration guides — either inline a manifest of the phase files (discovery, plan, implement, etc.) in SKILL.md's migration sections or hoist the phase files to a single index — so navigation stays one level deep.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is largely efficient — routing tables, terse troubleshooting entries, and pointer sections — but the maxTokens warning is repeated roughly six times (Critical Warnings, Converse section, Examples 1 and 7, ThrottlingException), the Harness paragraph packs five requirements into one dense run-on, and a few spots over-explain known concepts ('as IAM changes are eventually consistent'). This fits the 'efficient with minor trimmable over-explanation' anchor rather than the lean 5, but sits noticeably above the 3. | 4 / 5 |
Actionability | Guidance is copy-paste ready: 'aws bedrock-runtime converse --model-id us.anthropic.claude-sonnet-4-6 --messages … --inference-config '{"maxTokens":1024}'', 'aws bedrock list-foundation-models --region us-east-1', 'Config(retries={"max_attempts": 5, "mode": "adaptive"})', exact error signatures, and specific constraints like 'runtimeSessionId (≥33 chars)'. Common cases (invoke, throttling, RAG setup, agent deploy) are each covered by a concrete example. | 5 / 5 |
Workflow Clarity | Workflows are clearly sequenced with explicit validation: a Verify Dependencies gate (CLI presence, version, region access), the checkboxed invoke workflow (verify model access before invoking), the KB creation sequence ending in 'verify with Retrieve', Harness deployment with polling 'get-harness until status READY', and mandatory confirmation before destructive operations. The error-retry classification table and per-error troubleshooting entries provide feedback loops for recovery, satisfying the top anchor's checklists-and-validation criteria. | 5 / 5 |
Progressive Disclosure | The body is a well-signaled routing hub: a Table of Contents, capability tables mapping goals to references, and all 27 referenced reference files verified to exist on disk. However, the migration guides chain two levels deep (SKILL.md → migrate-knowledge-base-to-managed.md → references/bedrock-kb-migration/*.md with 9 phase files, and similarly for the AgentCore-harness migration with 5 files), which is a minor organization gap against the one-level-deep ideal. | 4 / 5 |
Total | 18 / 20 Passed |