How do I verify Codex actions with Provely?

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

Which surfaces does Codex use?

Which guidance files does Codex read?

Provely ships one Agent Skill for this host, so the agent knows to verify an outcome before it claims done. The renderer writes AGENTS.md fragment from one source. A new agent needs a profile, not core code.

How is a false completion claim blocked?

Guidance asks. A hook and the CI gate enforce. A host with neither still receives the verdict, and a person reads it before the work ships.

How do I connect Codex?

What else does the profile state?

• Verified 2026-09-12: Codex reads AGENTS.override.md, then AGENTS.md, at every level from the git root to the working directory. A deeper file wins. Source: https://learn.chatgpt.com/docs/agent-configuration/agents-md.
• Verified 2026-09-12: Codex reads [mcp_servers.<name>] from ~/.codex/config.toml and from .codex/config.toml of the project. A stdio entry states command, args, env and cwd. An HTTP entry states url. Source: https://learn.chatgpt.com/docs/extend/mcp.
• The Codex config is TOML, so this profile states no config_path and init writes no MCP file. Add [mcp_servers.provely] with command = "provely" and args = ["mcp"] by hand, or run codex mcp add.
• Verified 2026-09-12: a remote Codex MCP server takes auth, bearer_token_env_var and http_headers. The auth field defaults to oauth. Never write a key into the file. Source: https://learn.chatgpt.com/docs/extend/mcp.
• Verified 2026-09-12: Codex states 12 hook events, and this profile names all 12 in the order of the page. The earlier profile named none. Source: https://learn.chatgpt.com/docs/hooks.
• Verified 2026-09-12: hooks live in ~/.codex/hooks.json, in <repo>/.codex/hooks.json, or in a [[hooks.EventName]] table of config.toml. Source: https://learn.chatgpt.com/docs/hooks.
• Verified 2026-09-12: a project hook loads only when the .codex layer of the project is trusted. A fresh checkout therefore runs no hook. Source: https://learn.chatgpt.com/docs/hooks.
• Verified 2026-09-12: PreToolUse denies a tool call with permissionDecision deny, or with exit code 2. PermissionRequest denies with behavior deny. PostToolUse and UserPromptSubmit answer decision block. Source: https://learn.chatgpt.com/docs/hooks.
• Verified 2026-09-12: the Stop hook answers decision block with a reason, and Codex starts a continuation prompt that acts as a new user prompt. The hook does not end a turn. Source: https://learn.chatgpt.com/docs/hooks.
• Verified 2026-09-12: most hooks time out after 600 seconds. SessionEnd and Interrupt time out after 1 second, and 3 seconds is the maximum. Source: https://learn.chatgpt.com/docs/hooks.
• Verified 2026-09-12: requirements.toml holds allow_managed_hooks_only. A true value drops every user, project, session and plugin hook, and keeps the managed hooks. Source: https://learn.chatgpt.com/docs/hooks.
• This repository ships no Codex hook script, so hooks.supported stays verify_at_implementation. A yes makes the rendered files claim an enforcement that we do not ship. Read profiles/README.md.
• The CI gate and the scanner are the enforcement for Codex today. A PreToolUse hook is the next step, and it needs one script and one registration snippet.
• Needs verification: no page states the Codex version that added hooks. The supported range stays >=0.100.0. Check the installed binary before you register a hook.
• Cloud Codex tasks read the same AGENTS.md. The CI gate verifies before merge.

Codex keeps its action credentials, and no surface returns the verifier credentials to it. Read the credential boundary. An operation outlives the session that opened it. Read the API reference.