# How do I verify OpenAI Agents SDK actions with Provely?

> Connect OpenAI Agents SDK to Provely through the MCP server, the CLI, the SDK, the REST API. The guidance files, the hooks, and the CI gate.

Canonical: https://provely.sh/agents/openai-agents  
Last reviewed: 2026-09-05  

**OpenAI Agents SDK connects to Provely through the MCP server, the CLI, the SDK, the REST API. The agent does the action. Provely verifies the outcome and signs a receipt. OpenAI Agents SDK never holds the verifier credentials.**

A person verified the OpenAI Agents SDK capabilities on 2026-09-05. Agent products change monthly. Re-verify before you build.

Source: https://provely.sh/agents (retrieved 2026-09-05)

## Which surfaces does OpenAI Agents SDK use?

| Surface | Support | Detail |
| --- | --- | --- |
| [MCP server](/docs/mcp) | Supported | transport stdio, http |
| [CLI](/docs/cli) | Supported | Runs through the shell tool of the agent. |
| [SDK](/docs/sdk) | Supported | languages python, typescript; adapter `openai` |
| [REST API](/docs/api) | Supported | Any HTTP call reaches the control plane. |

## Which guidance files does OpenAI Agents SDK read?

The renderer writes prompt fragment, tool schemas from one source. A new agent needs a profile, not core code.

| Format | Path | Mode |
| --- | --- | --- |
| prompt fragment | `provely/prompt-fragment.md` | fragment |
| tool schemas | `provely/tool-schemas.json` | fragment |
| CI gate workflow | `.provely/ci/github-actions.yml` | fragment |

## How is a false completion claim blocked?

| Enforcement | Support | Detail |
| --- | --- | --- |
| Advisory guidance | Supported | The instruction files state the rules for every agent. |
| Lifecycle hooks | Supported | `output_guardrail` |
| CI gate | Supported | `provely verify --wait` fails the pipeline unless the verdict is VERIFIED. |

## How do I connect OpenAI Agents SDK?

### Connect OpenAI Agents SDK to Provely

1. **Install the CLI.** Run `npx provely --version`, or download the static binary.
2. **Register the MCP server.** Add the Provely server. Use `provely mcp` for stdio, or the hosted HTTP server with a bearer token. The tools are `begin`, `verify`, `status`, and `receipt`.
3. **Write the guidance files.** Run `provely init --agent openai-agents`. It writes `provely/prompt-fragment.md` (fragment), `provely/tool-schemas.json` (fragment), `.provely/ci/github-actions.yml` (fragment). The guidance carries one rule: a successful tool call is not completion.
4. **Install the hooks.** Enable the `output_guardrail` hooks. A hook calls the CLI and blocks a completion claim while an operation is PENDING.
5. **Add the CI gate.** Run `provely verify --wait <duration>` in the pipeline. The step fails unless the verdict is VERIFIED. Exit codes: 0 VERIFIED, 2 PENDING, 3 FAILED, 4 CONTRADICTED, 5 UNVERIFIABLE.
6. **Report the verdict exactly as returned.** VERIFIED: "The action is verified complete. Receipt: <id>." PENDING: "The action is accepted but not yet verified. Operation: <id>."

## What else does the profile state?

- Verified 2026-09-05: output guardrails run on the final agent output. `@output_guardrail` returns GuardrailFunctionOutput(output_info, tripwire_triggered). A tripwire raises OutputGuardrailTripwireTriggered. Source: https://github.com/openai/openai-agents-python (docs/guardrails.md).
- The hook event output_guardrail is code, not a script. The openai adapter (packages/mcp or the SDK) ships the guardrail that blocks a final answer that claims completion without a verdict object.
- Two version lines exist: openai-agents (Python) 0.22.0 and @openai/agents (JavaScript) 0.17.0 on 2026-09-05. The profile states one range that covers both. The adapter records the SDK version in effect per operation.
- Use the openai adapter. Wrap each side-effecting function tool so the action runs inside an operation. Attach the verdict to the tool result.

> OpenAI Agents SDK keeps its action credentials, and no surface returns the verifier credentials to it. Read [the credential boundary](/docs/security). An operation outlives the session that opened it. Read [the API reference](/docs/api).

### Which host versions does the profile cover?

`>=0.17.0`. The integrations team last verified the profile with openai-agents 0.22.0 (PyPI) and @openai/agents 0.17.0 (npm) on 2026-09-05.

## Read next

- [Verify Claude Code actions](https://provely.sh/agents/claude-code)
- [Verify Codex actions](https://provely.sh/agents/codex)
- [Verify Resend actions](https://provely.sh/verify/resend)
- [See every agent integration](https://provely.sh/agents)
- [Read the MCP server reference](https://provely.sh/docs/mcp)
- [Read the credential boundary](https://provely.sh/docs/security)