Agent not passing --agent when requesting --md output mode

[vmpn]

Right now if agent issues command like basecamp todos list --in <id> --md CLI will prompt to select todo set stopping the agent for human interaction.

Suspect this is due to how https://github.com/basecamp/basecamp-cli/blob/main/skills/basecamp/SKILL.md is phrased.

Could we get an option of env variables to tell basecamp to tell it should run in non interactive mode, without hoping LLM passes --agent flag. e.g. https://google-gemini.github.io/gemini-cli/docs/tools/shell.html#environment-variables

[robzolkos]

Fixed by #520 (merged to main).

You were right about the cause: --md doesn't suppress prompts. Only the machine-output flags (--agent/--json/--quiet/--ids-only/--count) did. So when the CLI runs under a PTY — where stdout looks like a terminal — and a target is ambiguous (e.g. a project with more than one todoset), it falls through to a blocking picker that wedges the agent.

The fix adds the escape hatch you asked for: BASECAMP_NONINTERACTIVE=1.

• Disables all interactive prompts (todoset / todolist / project / profile selection). They become actionable errors instead — e.g. Multiple todosets found — specify one with --todoset <id>.
• Does not change the output format, so --md still gives you Markdown. That's the key difference from --agent, which also forces quiet JSON.
• Read from the environment, so you set it once in your harness / shell / MCP config — the model never has to remember a per-call flag, which was your main concern (matching the pattern you linked from Gemini CLI).

The skill guidance is also updated to note that --md alone doesn't suppress prompts, and to point at the env var (and the disambiguating flags) as the fix.

This will be in the next release. Closing as resolved — thanks for the detailed report and the Gemini precedent.

[vmpn]

thank you for the enhancement