Bedrock
Bedrock boto3/CLI snippets, IAM model ARNs, and exception-by-exception debugging
Test report
- Verdict
- Works with setup
- Score
- Tested
- Jul 21, 2026
- Environment
- Claude Code 2.x (agent harness)
- Upstream re-checked
- Aug 10, 2026 · dcfb32b
Listed the repo tree via the GitHub API, fetched skills/bedrock/SKILL.md (393 lines, HTTP 200); spot-checked skills/bedrock/model-invocation.md (200) and .claude-plugin/marketplace.json (200, bedrock is listed) — no broken references and no curl|sh, base64 blobs, secret exfiltration or injection text, but SKILL.md never links model-invocation.md anywhere, so that companion file is orphaned and would never be loaded. Trigger phrasings I judged: SHOULD fire — "How do I call Claude on AWS Bedrock from boto3?", "Generate Titan text embeddings for my vector index", "I get AccessDeniedException invoking a foundation model in us-east-1"; SHOULD NOT — "Set up a SageMaker training job for my custom model", "Build a RAG pipeline over my docs with OpenAI embeddings and pgvector"; 5/5 correct, though the unqualified phrase "implementing RAG patterns" in the description is an over-trigger risk. Output test: I wrote a baseline Bedrock invocation snippet first (Converse API, botocore adaptive retries, bedrock vs bedrock-runtime note), then the skill-followed version — the skill added things I had not produced (foundation-model ARN format in a least-privilege IAM policy, `aws bedrock list-foundation-models --by-provider anthropic --query ...`, `iam simulate-principal-policy` as an access-denied probe, control-plane/data-plane command split), but pushed the retired model id anthropic.claude-3-sonnet-20240229-v1:0 and the legacy anthropic_version body instead of Converse, its retry example catches only ModelNotReadyException while the ThrottlingException section offers prose and no code, and one "Debug" block is fenced as python while containing a bash `aws` command. Frontmatter claims last_updated 2026-01-07 yet every model id is 2024-era with no mention of cross-region inference profiles, and the README's Codex example installs a nonexistent `rlhf` skill. No live AWS call was made; both artifacts were code, produced and compared in-context.
Scored on four weighted criteria — install, triggering, output vs. baseline, docs. How scoring works
- Installs cleanly 4/5
- Triggers reliably 5/5
- Output vs. baseline 4/10
- Docs & honesty 3/5
What Bedrock does
A reference skill for Amazon Bedrock that supplies boto3 and AWS CLI patterns for invoking foundation models, streaming responses, generating Titan embeddings, and holding multi-turn conversations, plus a control-plane vs data-plane CLI table, a least-privilege IAM policy with correct foundation-model ARNs, and troubleshooting sections keyed to AccessDeniedException, ModelNotReadyException, ThrottlingException and ValidationException. It triggers when you ask about calling foundation models on AWS, enabling model access, building embeddings or RAG on Bedrock, or debugging Bedrock API errors. Code examples still target the retired anthropic.claude-3-sonnet-20240229-v1:0 model id and the legacy invoke_model body format rather than the Converse API or cross-region inference profiles.
How to install Bedrock
/plugin marketplace add itsmostafa/aws-agent-skills
/plugin install aws-agent-skills
# Or single skill, manually:
git clone https://github.com/itsmostafa/aws-agent-skills /tmp/aws-agent-skills
cd /tmp/aws-agent-skills
mkdir -p ~/.claude/skills/bedrock
cp -R /tmp/aws-agent-skills/skills/bedrock/. ~/.claude/skills/bedrock/
Skills live in ~/.claude/skills/ (global) or .claude/skills/
(per-project). Restart Claude Code after installing.
Commands — how to trigger Bedrock
-
/bedrockBedrock boto3/CLI snippets, IAM model ARNs, and exception-by-exception debugging
It also activates on plain-language prompts like these:
-
Invoke a Bedrock foundation model for this task -
Build a RAG pipeline using Bedrock embeddings -
Configure model access for our Bedrock application
Frequently asked questions
- Is the Bedrock skill free?
- Yes. The skill itself is free from itsmostafa/aws-agent-skills. SkillProof publishes the install command and an independent test verdict at no cost.
- Does Bedrock work with Claude Code?
- We tested it with Claude Code 2.x (agent harness) on Jul 21, 2026. Verdict: Works with setup. Listed the repo tree via the GitHub API, fetched skills/bedrock/SKILL.md (393 lines, HTTP 200); spot-checked skills/bedrock/model-invocation.md (200) and .claude-plugin/marketplace.json (200, bedrock is listed) — no broken references and no curl|sh, base64 blobs, secret exfiltration or injection text, but SKILL.md never links model-invocation.md anywhere, so that companion file is orphaned and would never be loaded. Trigger phrasings I judged: SHOULD fire — "How do I call Claude on AWS Bedrock from boto3?", "Generate Titan text embeddings for my vector index", "I get AccessDeniedException invoking a foundation model in us-east-1"; SHOULD NOT — "Set up a SageMaker training job for my custom model", "Build a RAG pipeline over my docs with OpenAI embeddings and pgvector"; 5/5 correct, though the unqualified phrase "implementing RAG patterns" in the description is an over-trigger risk. Output test: I wrote a baseline Bedrock invocation snippet first (Converse API, botocore adaptive retries, bedrock vs bedrock-runtime note), then the skill-followed version — the skill added things I had not produced (foundation-model ARN format in a least-privilege IAM policy, `aws bedrock list-foundation-models --by-provider anthropic --query ...`, `iam simulate-principal-policy` as an access-denied probe, control-plane/data-plane command split), but pushed the retired model id anthropic.claude-3-sonnet-20240229-v1:0 and the legacy anthropic_version body instead of Converse, its retry example catches only ModelNotReadyException while the ThrottlingException section offers prose and no code, and one "Debug" block is fenced as python while containing a bash `aws` command. Frontmatter claims last_updated 2026-01-07 yet every model id is 2024-era with no mention of cross-region inference profiles, and the README's Codex example installs a nonexistent `rlhf` skill. No live AWS call was made; both artifacts were code, produced and compared in-context.
- What is the Bedrock SkillProof Score?
- 6.4/10 — installs cleanly 4/5, triggers reliably 5/5, output vs. baseline 4/10, docs & honesty 3/5.
- How do I install Bedrock?
- Copy the install command from this page, run it in your terminal, and restart Claude Code. Skills live in ~/.claude/skills/ (global) or .claude/skills/ inside a project.
- Can I use Bedrock with Cursor, Copilot, Gemini CLI, Codex or other AI tools?
- The SKILL.md format is native to Claude (Claude Code, Desktop, claude.ai). The instructions inside adapt to other assistants: Cursor rules, GitHub Copilot instructions, Windsurf rules, Custom GPTs, AGENTS.md for OpenAI Codex, and GEMINI.md for Google Gemini CLI — our conversion guides cover each, and the free converter on the tools page does the wrapping for you.