Command Parser

Overview

Run noisy commands through ~/.codex/skills/command-parser/scripts/command-parser.

The wrapper:

1. runs your command once,
2. writes logs into a temporary folder,
3. runs codex exec in that folder with read-only sandbox,
4. prints a compact summary from response.log,
5. removes the temp folder.

This keeps parser context focused on command output, even for very large logs.

Quick Start

• ~/.codex/skills/command-parser/scripts/command-parser <command...>
• COMMAND_PARSER_WARNINGS=1 ~/.codex/skills/command-parser/scripts/command-parser <command...>
• ~/.codex/skills/command-parser/scripts/command-parser --request-additional "What stack trace line caused failure?" <command...>

Recommended default:

• Do not request additional information unless absolutely necessary.
• Keep default compact extraction for speed and consistency.

Configuration

The script sources ~/.codex/skills/command-parser/.env.

Default values:

• COMMAND_PARSER_PROFILE=command-parser

Model and reasoning effort are not configured in .env.
They are read from the selected profile in ~/.codex/config.toml and remain the source of truth.
If the configured profile is missing from config.toml, the wrapper exits with an error.

codex exec invocation

The wrapper uses:

• -p <COMMAND_PARSER_PROFILE>
• -s read-only
• --skip-git-repo-check
• -o <tmp>/response.log
• stdout/stderr redirected to /dev/null

Output Expectations

• If no errors: No errors!
• Otherwise:
  • ## Errors with one bullet per distinct error
  • optional ## Warnings when COMMAND_PARSER_WARNINGS=1
• Optional: ## Requested Information only when an additional request is explicitly provided.
• Include file/line(/col) when present
• No advice, fixes, or extra commentary

Notes

• Avoid wrapping long-lived interactive/watch commands.
• For failed wrapped commands, the wrapper returns the wrapped command's exit code after printing the parsed summary.