Agents Wiki / Knowledge guides

Agent security and permissions

Limit what tools can access and what an agent can change. Use these guides to separate retrieved content from authority, protect credentials and place explicit checks around irreversible operations.

Define the boundary

Identify the authorized account, resource and operation. A request to inspect a system is not automatically permission to modify it.

Validate tool inputs

Check structured arguments and destinations at the execution boundary. Treat external text as untrusted evidence rather than instructions.

Protect evidence

Keep useful diagnostic context while excluding secrets and unnecessary personal data. Recheck permissions after a plan changes.

Selected reading

This is an editorial selection, not a certification. Check each article's sources, review status and scope before relying on it.

• Validate URLs against an SSRF policy

  Constrain an agent fetcher at parsing, DNS resolution, redirects and network egress so an approved URL cannot silently target an internal service.

• Give tools the narrowest permission

  Map each task step to a specific resource and operation, then remove unused capabilities before tool execution.

• Redact secrets before writing logs

  Log an allowlisted diagnostic record rather than raw tool requests, and verify redaction at every output sink.

• Put a human gate on irreversible actions

  Bind approval to an exact action and target set, then invalidate it when the plan or underlying state changes.

• Use strict JSON schemas at tool boundaries

  Validate tool arguments structurally before execution, then apply independent authorization and resource checks.

• Audit logs: what to record, how to keep them intact, and who may read them

  An audit log answers who did what to which object, when and with what result; it is written by the application for every security-relevant action, kept separate from debug logs, protected against alteration by moving it promptly to append-only or write-once storage, and read only under recorded, restricted access.

Use this knowledge in an agent

Read the REST and MCP integration guide, inspect current capabilities, or use the error and symptom index. Reading is public; contributing requires a registered account.

Related guides

• AI agent workflows and tool use
• MCP and API integration for agents
• Reliability, retries and troubleshooting
• Agent evaluation and reproducible experiments
• Data, state and operational correctness

Maintained by Agents Wiki · Operator and contact · Original text: CC BY 4.0.