fix(codex): inject dot-to-hyphen hook command note in Codex skills

[picklebento]

Summary

Hook commands in .specify/extensions.yml use dotted ids like speckit.git.commit, but Codex skills are named with hyphens (speckit-git-commit). The Claude integration handles this with an explicit instruction injected into each generated SKILL.md by ClaudeIntegration.post_process_skill_content. The Codex integration had no such override, so Codex would emit /speckit.git.commit (which does not resolve) instead of /speckit-git-commit.

This PR mirrors Claude's pattern in CodexIntegration:

• Adds _HOOK_COMMAND_NOTE and _inject_hook_command_note (identical regex / idempotency check to Claude's).
• Overrides post_process_skill_content to apply the note.
• Adds a small setup() override that re-reads each generated SKILL.md and applies post_process_skill_content (Claude already follows this same pattern).

Why not lift to SkillsIntegration base?

Vibe and Copilot are also affected by this latent bug, so a base-class lift would be a strictly better long-term fix. Keeping this PR focused on Codex per the contributing guideline ("Keep your change as focused as possible"); happy to follow up with a base-class refactor that deduplicates _inject_hook_command_note across Claude / Codex / Vibe / Copilot if maintainers prefer.

Test plan

• New TestCodexHookCommandNote class in tests/integrations/test_integration_codex.py mirrors TestClaudeHookCommandNote:
  • test_hook_note_injected_in_skills_with_hooks — full setup() produces a SKILL.md containing the note.
  • test_hook_note_not_in_skills_without_hooks — no false positives.
  • test_hook_note_idempotent — re-running injection is a no-op.
  • test_hook_note_preserves_indentation — note matches surrounding indentation.
• Updated test_non_claude_post_process_is_identity (which asserted on codex) to use agy instead, since agy is the closest remaining SkillsIntegration with no post_process_skill_content override.
• uv run pytest tests/ → 2866 passed, 34 skipped, 0 failures locally on Python 3.12.

Repro

Before this PR, against any project initialized with specify init . --integration codex --integration-options="--skills":

$ grep -c "replace dots" .agents/skills/speckit-specify/SKILL.md
0

After this PR:

$ grep -c "replace dots" .agents/skills/speckit-specify/SKILL.md
1

Discovered while shipping the Codex skills tree at instacart/carrot#787833 (Olive review comment surfaced the missing translation rule).

[Copilot]

Pull request overview

Fixes Codex skills generation so hook command IDs from .specify/extensions.yml (dot notation like speckit.git.commit) are correctly translated into Codex’s hyphenated skill slash-command names (/speckit-git-commit) by injecting an explicit “dot → hyphen” instruction into generated SKILL.md files—matching the existing Claude integration behavior.

Changes:

• Added hook-command normalization note injection + post_process_skill_content() override to CodexIntegration.
• Added a CodexIntegration.setup() post-pass to rewrite generated SKILL.md files with the injected note (mirrors Claude’s approach).
• Added Codex-focused tests for note injection behavior and updated the Claude test that asserts default SkillsIntegration post-processing is an identity function (now using agy).

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated no comments.

File	Description
tests/integrations/test_integration_codex.py	Adds tests validating dot-to-hyphen note injection, idempotency, and indentation preservation for Codex skills.
tests/integrations/test_integration_claude.py	Updates the “default post-process is identity” assertion to use agy instead of codex now that Codex post-processes skill content.
src/specify_cli/integrations/codex/init.py	Implements hook-note injection and a setup post-processing pass so generated Codex skills include the normalization instruction.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.