luigidemasi opened a new issue, #75: URL: https://github.com/apache/camel-kit/issues/75
> Imported from [luigidemasi/camel-kit#142](https://github.com/luigidemasi/camel-kit/issues/142), opened by [luigidemasi](https://github.com/luigidemasi) on 2026-07-11. > Closed by [luigidemasi/camel-kit#143](https://github.com/luigidemasi/camel-kit/pull/143), merged on 2026-07-12. --- ## Description Add first-class OpenAI Codex CLI support so users can initialize projects with: ```bash camel-kit init my-integration --ai codex camel kit init my-integration --ai codex ``` Reuse Camel Kit’s workflow and shared skills while generating Codex-native repository assets such as `AGENTS.md`, `.agents/skills/`, `.codex/config.toml`, and `.codex/agents/*.toml`. Do not generate deprecated or unsupported command and prompt scaffolding merely to fit the current agent abstraction. ## Expected Behavior - `--ai codex` is accepted by both init entry points and appears in help and diagnostics. - Generated workspaces contain a valid `AGENTS.md`. - All Camel Kit skills are installed under `.agents/skills/` and discoverable through `/skills` or `$camel-start`. - Codex-native MCP configuration includes the Camel, Camel Knowledge, and Citrus servers with resolved versions and appropriate tool allowlists. - Existing unrelated `.codex/config.toml` settings are preserved, or generation fails without changing the file. - Codex-specific dispatch guidance uses supported subagent capabilities while retaining an inline-execution fallback. - `camel-kit doctor` validates Codex TOML configuration and reports actionable errors. - No unused command directory is generated solely because command and skill locations are currently coupled. - Tests cover registry discovery, generation, skill references, MCP configuration, init behavior, doctor behavior, and post-init guidance. - An isolated Codex CLI smoke test verifies instruction, skill, custom-agent, and MCP discovery. - Existing AI targets continue to pass the full test suite. - CLI documentation and the Camel Kit website explain startup, skill invocation, MCP verification, repository trust, approvals, and sandboxing. ## Security Requirements - Do not modify the user’s global Codex configuration. - Do not manage OpenAI API keys or authentication. - Document that project configuration and hooks load only for trusted repositories. - Use least-privilege MCP and tool approval defaults. - Generate hooks only where their behavior has been verified against supported Codex lifecycle events. ## Additional Context The current generator couples command and skill directories, while MCP generation, doctor checks, and structural tests assume JSON. Codex uses repository-scoped skills and TOML project configuration, so these assumptions need to be generalized. Website impact: **required**, because this adds a public AI target, generated assets, setup workflow, and documented behavior. Official reference: https://developers.openai.com/codex/codex-manual.md ## Non-goals - OpenAI API or model integration. - Writing user-global configuration. - Speculative Codex assets unsupported by official project-level discovery. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
