docs: Claude Managed Agents Guard - #918
Conversation
Add a dual-language Guard page at /guards/claude-managed-agents/ with JS/Python tabs. Register the adapter on get-started and the agents quick start, and add one shared row on framework integrations. Co-authored-by: David Mytton <davidmytton@users.noreply.github.com>
Unpublished adapter snippets import packages this repo does not install. Also refresh llms.txt snapshots for the new Guard page. Co-authored-by: David Mytton <davidmytton@users.noreply.github.com>
List Claude Managed Agents on /agent-get-started/ as one product for both languages, with detect, install, and a link to the dedicated Guard page. Co-authored-by: David Mytton <davidmytton@users.noreply.github.com>
Pin Python installs to 40ea4896962a90a24cdbc4cfbfc80729c096da36, record anthropic>=0.92.0,<2, and document is_error on user.custom_tool_result. Leave the JavaScript SHA as ADAPTER_SHA. Co-authored-by: David Mytton <davidmytton@users.noreply.github.com>
Pin JS installs to cb35c8f92c3a2fb63fbeb9b386d79b1878c19d92 and record @anthropic-ai/sdk >=0.86.0 <1. Keep the Python pin at 40ea4896962a90a24cdbc4cfbfc80729c096da36. Co-authored-by: David Mytton <davidmytton@users.noreply.github.com>
Link examples/fastapi-claude-managed-agents-guard from the Claude Managed Agents Related list as the Python example. Co-authored-by: David Mytton <davidmytton@users.noreply.github.com>
Link examples/claude-managed-agents from Related and keep the Python FastAPI example cite. Co-authored-by: David Mytton <davidmytton@users.noreply.github.com>
|
Cursor Agent can help with this pull request. Just |
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
Keep Google ADK and Strands Agents Python from main, and add Claude Managed Agents to the shared listings and tests. Co-authored-by: David Mytton <davidmytton@users.noreply.github.com>
![arcjet-review[bot]](https://avatars.githubusercontent.com/u/24397786?s=60&v=4){kind=small}
There was a problem hiding this comment.
Arcjet Review — 🟢 Low Risk
Decision: Approved
Rationale: Documentation-only PR adding a new Claude Managed Agents Guard page and wiring it into related get-started, quick-start, and framework-integrations pages. Changes are additive: new MDX/snippet files, new entries in framework/SDK registries, matching Vercel redirects, and tests updated to cover the new keys. No production code, auth, secrets, or infrastructure changes. Snapshot tests are updated in lockstep with the source. .env examples use placeholder values (ajkey_yourkey, sk-yourkey).
Summary of Changes
Adds a new docs/guards/claude-managed-agents.mdx page (covering inbound user.message and custom-tool gating for Anthropic's hosted Managed Agents), plus JS+Python get-started and quick-start snippets. Wires the new framework keys (claude-managed-agents, claude-managed-agents-py) into prefs.ts, sdk.ts, sidebars, framework switcher/icon components, SdkReferenceLinkByFramework.astro, agent-get-started.mdx, get-started.mdx, guards/quick-start.mdx, guards/framework-integrations.mdx, guards/index.mdx, and reference/python.mdx. Updates tsconfig.json to exclude the new snippet dir from astro check, adds Vercel legacy-key redirects, adds screenshot/sdk test coverage, and refreshes the llms.txt snapshots.
Notes
Docs-only change; risk is low. Snapshot files under tests/llms-txt.test.ts-snapshots/ were updated in step with public/llms.txt / public/llms-full.txt, which is the expected pattern for this repo.
Review: d41d4177 | Model: anthropic/claude-opus-4-7 | Powered by Arcjet Review
Add email-provider placeholders, align wrap comments, move the agent-get-started install out of the table, and request one token per lookup. Co-authored-by: David Mytton <davidmytton@users.noreply.github.com>
The Integrations list is taller after adding Claude Managed Agents, so the home-light and home-dark snapshots no longer match. Co-authored-by: David Mytton <davidmytton@users.noreply.github.com>
Add Linux light/dark baselines for /guards/claude-managed-agents/ and update get-started, guards, guards-quick-start, and framework-integrations snapshots after the taller sidebar. Co-authored-by: David Mytton <davidmytton@users.noreply.github.com>
2 participants
Add a dedicated Claude Managed Agents Guard page (one slug, JS and Python) and wire it into get-started, guards/quick-start, and agent-get-started.
This is Anthropic’s hosted harness, not the Claude Agent SDK. Guard inbound
user.messagebefore send and custom tools onagent.custom_tool_use. Built-ins default toalways_allow. Custom-tool DENY isuser.custom_tool_resultwithis_error.Pins until publish:
@arcjet/guardcb35c8f9(@anthropic-ai/sdk>=0.86.0 <1)arcjet[claude-managed-agents]40ea4896(anthropic>=0.92.0,<2)Examples: claude-managed-agents, fastapi-claude-managed-agents-guard.