Repo-local .agents/skills skill is not injected into session Available skills on a fresh Codex CLI session

What version of Codex CLI is running?

0.116.0

What subscription do you have?

Pro

Which model were you using?

gpt-5.4 xhigh summaries auto fast

What platform is your computer?

Linux 6.17.0-19-generic x86_64 unknown

What terminal emulator and version are you using (if applicable)?

zsh 5.9 (x86_64-ubuntu-linux-gnu)

What issue are you seeing?

Summary

A repo-local skill at <repo-root>/.agents/skills/back-scrutiny/SKILL.md is not exposed to the model in a fresh session started from the repo
root, even though the Codex skills docs say .agents/skills should be discovered and symlinked skill folders are supported.

Docs: https://developers.openai.com/codex/skills

Environment

• Codex CLI version is 0.116.0
• Session was started from the repo root
• Session source is cli
• The skill file is present and has the expected frontmatter:
• name: back-scrutiny
• description: ...
• The repo is trusted in ~/.codex/config.toml
• There is no [[skills.config]] disable section affecting this skill

What steps can reproduce the bug?

1. Start a fresh Codex CLI session from <repo-root>.
2. Ensure <repo-root>/.agents/skills/back-scrutiny exists and resolves to a valid SKILL.md.
3. Ask Codex to use back-scrutiny.
4. Inspect the session bootstrap metadata.

~/.codex/config.toml

  model = "gpt-5.4"
  model_reasoning_effort = "high"
  personality = "pragmatic"
  web_search = "live"
  service_tier = "fast"

  [projects."<repo-root>"]
  trust_level = "trusted"

  [features]
  unified_exec = true
  steer = true
  multi_agent = true

  Notes:

  - No [[skills.config]] section is present.
  - No skill-specific disable rule is present.
  - The repo is marked trusted.

<repo-root>/.agents/skills/back-scrutiny/SKILL.md

---
  name: back-scrutiny
  description: Backend code scrutiny and review. Use this skill for reviewing backend code quality, identifying issues, and suggesting
  improvements. Python emphasis.
  ---

Skill Path Metadata

<repo-root>/.agents/skills/back-scrutiny
-> <sibling-repo-root>/.agents/skills/back-scrutiny

Notes:

• The repo-local skill is exposed via symlink.
• The target directory contains SKILL.md.
• No BOM was present at the start of SKILL.md; the file begins with ---.

Minimal File Inventory

<repo-root>/
.agents/
skills/
back-scrutiny -> <sibling-repo-root>/.agents/skills/back-scrutiny

What is the expected behavior?

The skill should be included in the session’s discovered Available skills set and be invokable by name.

Additional information

• The skill appears structurally valid; I do not see missing required metadata or companion files.
• Other skills available in the same client are symlinked, so this does not look like a generic symlink failure.
• This looks like a repo-local .agents/skills discovery or prompt-injection bug, or a mismatch between the docs and CLI behavior.