Bug: script runner misidentifies 'codex' in model names as the codex runtime binary

[ryan-niemes-helix]

Description

When a script command includes --model gpt-5.3-codex (or any model name containing the substring codex), the APM script runner incorrectly transforms the copilot command into a codex exec command, causing execution to fail with No such file or directory: 'codex'.

Steps to Reproduce

1. Add a script to apm.yml:

   scripts:
     fix-issue: "copilot --allow-all-tools --model gpt-5.3-codex -p fix-issue.prompt.md"

2. Run: apm run fix-issue

Result: APM transforms the command to codex exec --model gpt-5.3-codex -p and fails:

Script execution error: [Errno 2] No such file or directory: 'codex'

Expected: APM runs copilot --allow-all-tools --model gpt-5.3-codex -p <content>

Root Cause

In src/apm_cli/core/script_runner.py, _transform_runtime_command uses:

if re.search(r"codex\s+.*" + re.escape(prompt_file), command):

This regex matches codex anywhere in the command string — including as a substring of --model gpt-5.3-codex. Since gpt-5.3-codex --allow-all-tools ... fix-issue.prompt.md matches codex\s+.*fix-issue.prompt.md, the entire command gets rewritten as a codex exec invocation.

Suggested Fix

Anchor the match so codex is only recognized as a standalone word at the start of the runtime binary, not inside a flag value:

if re.search(r"(^|\s)codex\s+.*" + re.escape(prompt_file), command):

Workaround

Omit --model entirely and let Copilot CLI default to its built-in model selection:

scripts:
  fix-issue: "copilot --allow-all-tools -p fix-issue.prompt.md"

Environment

• APM version: 0.8.3
• Platform: macOS