Conciseness

The body is an efficient router — tables, decision guides, and pointers rather than concept explanations Claude already knows. However, the maxTokens/ThrottlingException warning is repeated in Critical Warnings, the Converse section, Throttleshooting, and Examples 1 and 7, and 'You MUST read [X] before responding' is restated nearly verbatim across many sections. This is deliberate emphasis but is trimming-able, matching 'Efficient; minor instances of over-explanation that could be trimmed' rather than the lean anchor 5.

Actionability

Concrete, copy-paste-ready commands cover the common cases: `aws bedrock list-foundation-models --region us-east-1`, the full Example 4 converse command with model ID and `--inference-config '{"maxTokens":1024}'`, quota-check commands, retry config `Config(retries={"max_attempts": 5, "mode": "adaptive"})`, and exact error names mapped to specific fixes (e.g., the double-underscore action-group name pattern, `prepare-agent` after config changes).

Workflow Clarity

Workflows are sequenced with most checkpoints present: the Invoke-a-model checklist, KB create → connector → ingest → 'verify with Retrieve', 'polling get-harness until status READY', and 'Run prepare-agent after ANY configuration change'. It falls short of anchor 5 because several multi-step procedures (agents, guardrails, AgentCore deploy) delegate their detailed sequence and validation checkpoints to reference files rather than showing feedback loops inline — the body states the gates exist but the fix-and-retry loops themselves live one level down.

Progressive Disclosure

The body is a well-signaled overview with one-level-deep references (all 25 referenced `references/*.md` paths and `scripts/fetch_bedrock_agent.py` verified to exist), a TOC, and routing tables per capability. Scored against the actual bundle structure, it does not quite reach anchor 5 because two areas nest a second level: `migrate-bedrock-agents-to-agentcore-harness.md` links into `references/bedrock-agents-to-agentcore-harness/{cli,deploy,discovery,eligibility,mapping}.md`, and `agentcore-payments.md` links to `agentcore-payments-setup-script.md` and `agentcore-payments-wiring.md`. Good structure overall with minor organization gaps.

Specificity

Lists multiple specific concrete actions with comprehensive coverage: 'model invocation (Converse API, InvokeModel), RAG with Knowledge Bases, Bedrock Agents, Guardrails, and AgentCore', plus migration, troubleshooting, prompt caching, and cost tracking. It matches the anchor 'Lists multiple specific concrete actions; comprehensive coverage' and stays in third person ('Builds', 'Covers', 'Applies').

Completeness

Explicitly answers both: what ('Builds generative AI applications on Amazon Bedrock. Covers...') and when ('Applies when invoking models, setting up Knowledge Bases, creating agents...'), with concrete trigger phrases and even negative triggers ('NOT for custom model training, Rekognition, or Comprehend'). This matches the anchor-5 example structure exactly.

Trigger Term Quality

Comprehensive natural-term coverage including explicit synonyms users would actually say: 'migrating/porting/converting', 'ThrottlingException, AccessDeniedException', 'prompt caching', 'quota and throttling diagnosis', 'cost tracking', 'choosing models (Claude, Llama, Nova, Titan)'. It includes error names and task phrasings a user would naturally utter.

Distinctiveness Conflict Risk

Clear niche (Amazon Bedrock and its named sub-services) with distinct triggers and explicit exclusions ('NOT for custom model training, Rekognition, or Comprehend') that prevent it firing for adjacent AWS AI skills. Minimal conflict risk.