Interactive Codex rejects skill-style slash commands like /ralph-loop instead of routing them through skill injection

Problem

In interactive Codex, entering a skill-style slash command such as /ralph-loop is rejected as an unknown built-in command instead of being passed through whatever skill/command injection layer is available.

Current error:

Unrecognized command '/ralph-loop'. Type "/" for a list of supported commands.

This makes skill invocation ergonomics inconsistent with environments where users naturally start a skill by typing its slash name.

Current Behavior

• Environment: Linux x86_64
• Codex version: codex-cli 0.115.0
• In an interactive Codex session, typing /ralph-loop "..." --max-iterations 20 ... fails immediately with:
• Unrecognized command '/ralph-loop'. Type "/" for a list of supported commands.
• Codex appears to only accept a fixed built-in slash-command list, with no fallback path for skills/custom slash commands.

Expected Behavior

One of these should happen instead:

1. Skill-aware behavior:

• If /ralph-loop maps to an installed skill/command, Codex should route it through the skill injection/dispatch layer.

1. Explicit fallback behavior:

• If Codex does not support slash-prefixed skill invocation, it should provide a clearer message explaining that only built-in slash commands are supported and how to invoke installed skills in Codex.

Why This Matters

• Users naturally try /skill-name when a skill exists.
• The current failure looks like a parser limitation rather than an intentional product boundary.
• This blocks workflows that are documented or habituated around slash-style skill invocation.

Reproduction

1. Start an interactive Codex session.
2. Enter:

/ralph-loop "Continue fixing Rosetta issues discovered by Molly's testing..." --max-iterations 20 --completion-promise "At least 2 more Rosetta gaps fixed..."

1. Observe:

Unrecognized command '/ralph-loop'. Type "/" for a list of supported commands.

Expected Fix Direction

• Add a post-parse fallback path for unknown slash commands so installed skills/custom commands can be resolved before showing the generic "Unrecognized command" error.
• If this is intentionally unsupported, document the limitation and surface a targeted error message with the supported skill invocation syntax for Codex.

Related

• Possibly adjacent to slash-command handling gaps such as openai/codex#3641, but this report is specifically about interactive Codex rejecting skill-style slash commands like /ralph-loop.