codex - 💡(How to fix) Fix Expose explicit and implicit skill metadata in Stop hook payloads [1 participants]

What variant of Codex are you using?

CLI

What feature would you like to see?

Expose canonical skill invocation metadata on the existing Stop hook payload.

Today Codex already knows about two kinds of skill usage:

• explicit skill usage from selected/mentioned skills
• implicit skill usage detected from shell commands

But repo-local hooks cannot consume that information directly. Downstream integrations currently have to either:

• parse rollout transcripts to recover explicit skill usage, or
• reimplement Codex's shell-based implicit skill detection heuristically

It would be useful if the Stop hook input included two additive fields such as:

• mentioned_skills: [{ name, path }]
• implicit_skills: [{ name, path }]

This would let integrations consume Codex's own canonical view of skill usage from a stable hook surface, without depending on analytics-only internals or transcript parsing.

Why Stop seems like the right place:

• Codex already has the turn-level skill context by then
• the payload is already used for end-of-turn integration behavior
• the change is additive, so existing Stop hooks would continue to work unchanged

Additional information

I looked through the current CLI behavior and found:

• explicit skill information exists during turn construction
• implicit skill invocation is already detected internally
• neither is currently exposed to hook consumers in a first-class way

I have a focused implementation ready on a fork if this approach aligns with the intended direction:

• branch: feat/stop-hook-skill-metadata

The patch does the following:

• adds mentioned_skills and implicit_skills to the existing Stop hook input
• populates them from the existing explicit and implicit skill paths
• updates the generated stop-hook schema fixture
• adds focused tests for both explicit and implicit cases

If this direction is acceptable, I'd be happy to open an invited PR.