Creating a skill + automation does not trigger immediate execution
Resolved 💬 2 comments Opened Feb 20, 2026 by twentyOne2x Closed Feb 21, 2026
💡 Likely answer: A maintainer (github-actions[bot], contributor)
responded on this thread — see the highlighted reply below.
Summary
Codex did not reliably complete an explicitly ordered, high-autonomy workflow end-to-end in one run:
- create/update a custom skill,
- create/update an automation that uses that skill,
- start execution immediately in the current session.
Expected Behavior
- When the user gives an explicit sequence (setup first, then execute now), Codex should:
- complete all setup steps,
- validate setup artifacts,
- immediately start execution without requiring additional prompts.
Actual Behavior
- The setup was only partially usable after the first pass.
- Extra follow-up prompts were required before behavior matched intent.
- In this case, custom skill discoverability/autocomplete did not work initially, which blocked the intended “create + run now” flow.
Core Product Gap
There appears to be a gap between:
- successful artifact creation on disk, and
- runtime readiness/activation (discoverability + immediate execution path).
For unattended workflows, this gap is critical because the user expects a single-prompt handoff.
Suggested Fixes
- Add strict ordered-execution mode for prompts that clearly specify setup-then-run sequencing.
- Add post-creation readiness checks for skill/automation flows (parse, registration, discoverability).
- If readiness checks fail, auto-remediate in the same run or return an explicit blocking failure with a next action.
- Add an option to execute newly created automation immediately (one-shot run), not only schedule it.
Environment
- CLI/App version observed:
codex-cli 0.104.0-alpha.1 - OS: macOS 26.3 (Darwin 25.3.0 arm64)
2 Comments
Potential duplicates detected. Please review them and close your issue if it is a duplicate.
Powered by Codex Action
This is by design currently, not a bug. New skills are detected by a file watcher. There's a 10-second debounce to allow a new skill and its related files to be written to disk before it is recognized by the harness. If you want it to be recognized in an automated flow, you'll need to tell the agent to wait.