OpenClaw scheduled agent task fails on approval policy while the same command works interactively
Grimmoner · reddit · 2026-09-14
- A self-hosting OpenClaw user has an hourly scheduled agent job running a custom CLI tool: the exact same command succeeds every time when triggered manually in chat, but automated runs get denied by the exec approval policy, citing headless runs being unable to wait for interactive approval.
- Ruled out: allowlist grants (correct and effective interactively), approval fallback tweaks (turns instant fail into a 10-minute hang), rewriting agent instructions to read results from files, moving the command to a plain scheduled script, and a helper meant to auto-approve requests (connected but never saw any).
- The strangest part is inconsistency: back-to-back runs with identical steps sometimes pass and sometimes fail, with no visible difference.
- The author concludes it's an approval-layer issue rather than data breakage and is asking the community for similar "works interactively, fails when scheduled" cases.
More from coding & agent
- Benzi lets you chat with any GitHub repo instantly — DonkeyTheKing · 2026-09-14
- Automating ComfyUI with Claude Code + MCP: local ad pipeline on RTX 4060 Ti — EndPsychological8822 · 2026-09-14
- First time 'rolling off' one AI coding harness and handing the codebase to another — arieljalali · 2026-09-14
- Rebuilding marble maze parts from photos with Astra, then making a three.js physics builder — OfirPress · 2026-09-14
- Developer builds tool that puts ChatGPT, Claude and Gemini in one debate to cross-check each other — medredac · 2026-09-14
- Dev builds a game in 5 hours with local Qwen3.8 27B on an overclocked RTX 3090, plans to open source — EcstaticDentist · 2026-09-14