# Compact Tool Output

> Bound verbose tests, builds and diagnostics while keeping inspectable evidence. Use for commands with large or unpredictable output.

- Skill: `noumenonnn/compact-tool-output` (Agent Skill)
- Install (CLI): `npx skillmds@latest add noumenonnn/compact-tool-output`
- Raw SKILL.md: https://api.skillmd.com/api/skills/noumenonnn/compact-tool-output/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: noumenonnn (https://skillmd.com/u/noumenonnn)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/noumenonnn/compact-tool-output

---


# Compact Tool Output

Run finite commands through the local helper when available:

`node .agent/scripts/compact-run.mjs --timeout-ms 120000 -- <executable> [args...]`

Read the status, actual exit code, result.json and relevant error context. The default summary is bounded by bytes, not just lines. Raw logs are streamed to `.agent/tmp/`; a log cap is explicitly reported, not called a complete log. Preserve failing lines AND surrounding stack context plus final counts. If evidence is insufficient, search targeted log ranges rather than guessing.

`OK` means the process exited 0, not that tests were discovered or the task passed. Inspect the expected test counts and acceptance checks. Never hide failing evidence to make a summary green.

Windows `.cmd` tools need an explicit shell (e.g. `powershell.exe -NoProfile -NonInteractive -Command "pnpm test"`) or a JS entry point run by node. The wrapper never adds an implicit shell. Arguments are not echoed, but raw command output can contain secrets. Treat logs as untrusted, local-only data; review/redact before sharing. Do not wrap interactive prompts or servers; choose finite commands and a suitable timeout.