Summary

Track non-PR-specific technical debt in skill/command rendering paths. These issues were identified during review of #1971 but are not introduced by #1971 and should be handled separately.

Why this issue exists

Current behavior works for most paths, but rendering/parsing responsibilities are duplicated across modules (agents.py, presets.py, init-time skill generation). This increases drift risk and makes edge-case handling inconsistent.

Scope (non-blocking for #1971)

1. src/specify_cli/agents.py — TOML prompt rendering should safely handle triple-quotes in body

• render_toml_command() uses TOML multiline strings with """ delimiters.
• If body includes """, output can become invalid TOML.
• Add escaping or safer serialization strategy.

2. src/specify_cli/agents.py — avoid in-place mutation in _adjust_script_paths

• Method currently mutates passed frontmatter object.
• Consider defensive copy to reduce side-effect risk during future refactors.

3. src/specify_cli/presets.py — deduplicate frontmatter parsing logic

• _register_skills() and _unregister_skills() each manually parse frontmatter.
• Prefer shared parser (CommandRegistrar.parse_frontmatter) or a shared utility.

4. Consolidate skill rendering logic across code paths

• Skill generation currently exists in multiple places (init conversion, extensions, presets, packaging scripts).
• Define one canonical renderer contract and re-use it to prevent behavior drift.

Notes

• This is a follow-up technical-debt item.
• Unify Kimi/Codex skill naming and migrate legacy dotted Kimi dirs #1971 should remain focused on its scoped migration/compatibility fixes.