Concise output mode for run-many: Suppress logs for successful tasks (AI Agent friendly)

[YKDZ]

Hi, 👋

I'm currently working in a large monorepo and looking for a way to streamline the terminal output when running nx run-many or tasks with multiple dependencies.

The Problem:
Currently, when running multiple tasks, the terminal is flooded with the standard output of all involved tasks.

What I've Tried:
I have reviewed the CLI documentation and tested the --outputStyle flag, but unfortunately, none of the existing options solve the issue for my specific use case:

• dynamic / dynamic-legacy: While these look perfectly clean to a human developer in a terminal, they rely heavily on ANSI escape sequences to overwrite console lines. For an AI Agent reading a raw stdout text stream, these escape characters cause severe context pollution and hallucination.
• static: This provides clean plain text, but it completely dumps the full logs of every successful task sequentially, causing massive token bloat.
• stream: Similar to static, it streams everything without filtering.

The Ideal Behavior:
I am looking for a strictly "Failures-Only" plain text mode. The ideal behavior would be:

• If a task succeeds: Show absolutely no stdout, just a single line indicator (e.g., ✔ task:build (success)).
• If a task fails: Print the full standard error / stack trace only for the failed task.

The Use Case & Impact:
Beyond developer terminal readability, my primary pain point involves AI Agents. I am utilizing LLM-based agents to assist with CI/CD and automated monorepo maintenance.

When an agent executes an Nx command and receives thousands of lines of successful build logs:

1. Context Window Bloat: It completely fills up the LLM's context window with useless noise, wasting tokens and hitting limits.
2. Attention Shift: The agent often gets distracted by harmless warnings or verbose outputs from successful tasks, ultimately missing the actual error at the bottom.

My Questions:

1. Is there currently an undocumented combination of flags, a plugin, or a specific custom logging configuration to achieve this "quiet on success, loud on failure" plain text mode?
2. If not, could this be considered as a Feature Request? A flag like --silent-success or --errors-only would be a massive quality-of-life improvement for both CI environments and modern AI-driven workflows.

Thanks in advance for your insights!