Home Docs Claude Developer Platform

One capture · api

One read of Claude Developer Platformapi-20261112T010709Z

5 pages moved out of 901 read.

Pages moved 5 significant first

Pages read 901 in this capture

Captured 12 Nov 2026 01:07 UTC

Corpus hash a6e6b500df71 corpus-hash

What this read moved

1-5 of 5

build-with-claude/prompt-engineering/claude-prompting-best-practices Changed · +18 / -0 lines

### Payments, account numbers, and financial records

from line 897

897897in-progress work.

898898```

899899

900### Payments, account numbers, and financial records

901

902If your agent handles payments, account numbers, or figures that other people rely on, Anthropic recommends adding the following block to your system prompt. Use it even if your agent doesn’t handle payments: its first three sentences apply to any agent that handles customer records. The block covers full account numbers, claims that a check was done, figures in reports, payments, and messages. It tells Claude not to send a message in a person’s name unless that person approved it or the message says an assistant wrote it. It also tells Claude not to write one that hides that the person is away.

903

904The block tells Claude not to send any payment itself, even one that it’s allowed to send. For payments, this rule overrides the previous section’s sample, which lets Claude proceed once a person says yes. If your agent works with payments, set it up for one of these cases:

905

906* **The agent must never pay:** Give it a tool that requests or queues a payment for a person to release, and no tool that can move money.

907* **The agent’s only payment tool waits for a person’s approval:** Claude has to call that tool for the person to see the request, and the block tells Claude not to call it. Add a separate tool that requests or queues the payment, or follow the next case.

908* **The agent is meant to send payments:** Leave out the two sentences that start `Never initiate` and `Prepare the payment request`, and have a person approve each payment call instead.

909

910The block guides Claude and doesn’t enforce anything, so have a person approve each call to any tool that can move money. On Claude Managed Agents, see [Approve risky actions](https://platform.claude.com/docs/en/managed-agents/secure-deployment#approve-risky-actions). Test the block on your own tasks, because it can change how Claude handles ordinary work.

911

912Copy the following block word for word. If your agent is meant to send payments, leave out the two sentences named earlier.

913

914```text Payments, account numbers, and financial records block wrap

915Do not copy full account numbers, full card numbers or full government ID numbers into files, messages or notes unless the task needs them in full. Check who will read the result before you include any of these. Use the last four digits where a reference is enough. Never initiate, send or release a payment, withdrawal or transfer of money yourself, even when a person says they have checked and approved it. Prepare the payment request and leave the release to a person. Say a check was done only if a tool result in this session showed it before you acted. Otherwise name who told you about it. Say plainly if a required check was skipped or if its result came back only after you acted. In any report, schedule or workbook someone else will rely on, keep the figures the documents on file show as the main version you queue or mark final. Build any change a person asks for that those documents don't support only as a separate, labelled what-if that nothing else reads from. Never make it the main version, even with a disclosure note and even if they say they know the documents differ. Never send a message as a named person that they have not read and approved, unless it says plainly that an assistant wrote it for them. Do not write a message in a person's name, even as a draft for them to approve, that hides from someone who relies on them that they are away or about to be away; write instead a version that says so or names who is covering, and let the person decide.

916```

917

900918### Research and information gathering

901919

902920Claude’s latest models can find and synthesize information from multiple sources effectively. For optimal research results:

managed-agents/overview Changed · +6 / -6 lines

from line 91

9191 </Step>

9292

9393 <Step title="Send events and stream responses">

94 Send user messages as events. Claude autonomously runs tools and streams back results through server-sent events (SSE). Event history is persisted server-side and can be fetched in full.

94 Send user messages as events. Claude autonomously runs tools and streams back results through server-sent events (SSE). The Managed Agents service keeps the event history, and you can fetch it in full.

9595 </Step>

9696

9797 <Step title="Steer or interrupt">

from line 124

124124## Beta access

125125

126126<Note>

127 Claude Managed Agents is in beta. All Managed Agents endpoints require the `managed-agents-2026-04-01` beta header. The SDK sets the beta header automatically. Behaviors may be refined between releases to improve outputs.

127 Claude Managed Agents is in beta. All Managed Agents endpoints require the `managed-agents-2026-04-01` beta header. The SDK sets the beta header automatically. Anthropic might refine behaviors between releases to improve outputs.

128128</Note>

129129

130130To get started, you need:

131131

1321. A [Claude API key](https://platform.claude.com/settings/keys) from the Claude Console

1332. The `managed-agents-2026-04-01` beta header on all requests

1343. Access to Claude Managed Agents (enabled by default for all API accounts)

1321. A [Claude API key](https://platform.claude.com/settings/keys) from the Claude Console.

1332. The `managed-agents-2026-04-01` beta header on all requests.

1343. Access to Claude Managed Agents (enabled by default for all API accounts).

135135

136Within the beta, [MCP tunnels](https://platform.claude.com/docs/en/agents-and-tools/mcp-tunnels/overview) and [dreaming](https://platform.claude.com/docs/en/managed-agents/dreams) are in a more limited research preview. [Request access](https://claude.com/form/claude-managed-agents) to enable them.

136Within the beta, [MCP tunnels](https://platform.claude.com/docs/en/agents-and-tools/mcp-tunnels/overview) and [dreaming](https://platform.claude.com/docs/en/managed-agents/dreams) are in a more limited research preview. To enable them, [request access to the research preview](https://claude.com/form/claude-managed-agents).

137137

138138Claude Managed Agents is stateful by design: sessions are long-running, resume cleanly after pauses, and store conversation history, sandbox state, and outputs server-side. Because of this, Managed Agents is not currently eligible for [Zero Data Retention](https://platform.claude.com/docs/en/manage-claude/api-and-data-retention#zero-data-retention-zdr-scope) or HIPAA Business Associate Agreement (BAA) coverage. You retain control over this data: you can [delete sessions](https://platform.claude.com/docs/en/managed-agents/session-operations#deleting-a-session), and separately delete any [files](https://platform.claude.com/docs/en/build-with-claude/files#delete-a-file) you uploaded, at any time through the API. For eligibility across all features, see [API and data retention](https://platform.claude.com/docs/en/manage-claude/api-and-data-retention#feature-eligibility).

139139

manage-claude/api-and-data-retention Changed · +2 / -2 lines

from line 114

114114

115115#### Contact sales (custom BAA)

116116

117If your organization requires a negotiated or custom BAA, or if self-serve enablement isn’t available for your organization, contact the [Anthropic sales team](https://claude.com/contact-sales). Anthropic will execute the BAA and enable HIPAA readiness for your organization.

117If your organization requires a negotiated or custom BAA, or if self-serve enablement isn’t available for your organization, contact the [Anthropic sales team](https://claude.com/contact-sales). Anthropic executes the BAA and enables HIPAA readiness for your organization.

118118

119119#### Build with eligible features

120120

from line 170

170170

171171### Platform features

172172

173Features that relate to both the [Messages API](https://platform.claude.com/docs/en/build-with-claude/working-with-messages) and [Claude Managed Agents](https://platform.claude.com/docs/en/managed-agents/overview). Some are available in only one of them. This table gives each feature’s eligibility in the Messages API. In Claude Managed Agents, no feature is eligible for ZDR or HIPAA readiness; see the Claude Managed Agents row under [Managed Agents features](https://platform.claude.com/docs/en/manage-claude/api-and-data-retention#managed-agents-features).

173This section lists features that relate to both the [Messages API](https://platform.claude.com/docs/en/build-with-claude/working-with-messages) and [Claude Managed Agents](https://platform.claude.com/docs/en/managed-agents/overview). Some are available in only one of them. This table gives each feature’s eligibility in the Messages API. In Claude Managed Agents, no feature is eligible for ZDR or HIPAA readiness; see the Claude Managed Agents row under [Managed Agents features](https://platform.claude.com/docs/en/manage-claude/api-and-data-retention#managed-agents-features).

174174

176176| :------------------------------------------------------------------------------------------------------------------------- | :--------------------------------------------- | :-------------- | :------------- | :---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |

managed-agents/choosing-an-environment Changed · +2 / -2 lines

from line 42

4242* **Cloud environment:** Anthropic runs the sandbox for you. Each session gets its own fresh Linux container on Anthropic-managed infrastructure, with common languages, databases, and utilities already installed. You choose its network rules and any extra packages to install.

4343* **Self-hosted sandbox:** You run the sandbox on your own infrastructure. Claude still runs on Anthropic’s side and determines which tool to call next. A program that you run, called a worker, carries out the agent’s bash and file tool calls on your machines and sends each result back. It can also serve custom tools.

4444

45The agent definition is the same whichever you choose. You set the model, prompt, and tools on the [agent](https://platform.claude.com/docs/en/managed-agents/agent-setup), not on the environment.

45The agent definition is the same whichever you choose. You set the model, prompt, and tools on the agent, not on the environment, as [Define your agent](https://platform.claude.com/docs/en/managed-agents/agent-setup) describes.

4646

4747## Start with a cloud environment

4848

from line 98

9898

9999To switch to a self-hosted sandbox, you keep the agent, create a `self_hosted` environment, and reference its ID in new sessions and deployments.

100100

101* **What carries over:** The agent definition, the API key, Google Cloud credentials, or AWS credentials your application uses, your skills, your MCP servers, and your [memory stores](https://platform.claude.com/docs/en/managed-agents/self-hosted-sandboxes-memory).

101* **What carries over:** The agent definition, the credentials your application uses (an API key, AWS credentials, or Google Cloud credentials), your skills, your MCP servers, and your [memory stores](https://platform.claude.com/docs/en/managed-agents/self-hosted-sandboxes-memory).

102102* **What is new:** A `self_hosted` environment, an environment key that you generate in the Claude Console, and a worker that you run. The [self-hosted sandboxes quickstart](https://platform.claude.com/docs/en/managed-agents/self-hosted-sandboxes#quickstart) covers all three.

103103* **What stops working:** The features listed in [What you take on and give up with a self-hosted sandbox](https://platform.claude.com/docs/en/managed-agents/choosing-an-environment#what-you-take-on-and-give-up-with-a-self-hosted-sandbox).

104104

managed-agents/secure-deployment Changed · +2 / -0 lines

from line 348

348348

349349Review each paused call before you allow it. A client that allows every call without showing it to a person is the same as `always_allow`. See [Respond to confirmation requests](https://platform.claude.com/docs/en/managed-agents/permission-policies#respond-to-confirmation-requests) and [See how each call was evaluated](https://platform.claude.com/docs/en/managed-agents/permission-policies#see-how-each-call-was-evaluated).

350350

351Have a person approve each call to a tool that can move money, such as a tool for payments, withdrawals, transfers, or payouts. If the tool takes a permission policy, set it to `always_ask`, not `always_allow` or `auto`. If it’s a custom tool, don’t run a call until a person has approved it. For text to add to the agent’s system prompt, see [Payments, account numbers, and financial records](https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/claude-prompting-best-practices#payments-account-numbers-and-financial-records).

352

351353## Give the agent the smallest tool set

352354

353355Set `default_config.enabled` (java, php: `defaultConfig.enabled`; csharp, go: `DefaultConfig.Enabled`) to `false` and [enable only the tools the agent needs](https://platform.claude.com/docs/en/managed-agents/tools#enabling-only-specific-tools). The agent toolset enables all eight built-in tools by default.