Feature: PreToolUse/PostToolUse hooks for local slash command (skill) execution

Summary

When a skill is invoked via a CLI slash command (e.g. /issue, /commit), the skill prompt is injected directly into the conversation context. This bypasses the Skill tool entirely, so PreToolUse[Skill] hooks never fire.

This makes it impossible to intercept or augment skill execution at the hook level when skills are triggered the most natural way — as slash commands.

Current Behavior

• /issue → skill prompt injected as system context → Skill tool never called → PreToolUse[Skill] hook does not fire
• Skill tool called mid-conversation → PreToolUse[Skill] hook fires correctly

Desired Behavior

A hook event (e.g. PreSkillUse / PostSkillUse, or PreCommand / PostCommand) that fires when a skill is invoked via a local slash command, with the skill name and arguments available on stdin — consistent with the existing hook contract.

Use Case

Project-level persona agents (subagents in .claude/agents/) configured to evaluate feature proposals before a GitHub issue is created. The evaluation is wired as a PreToolUse[Skill] hook targeting the issue skill, but it only works when the skill is called programmatically mid-conversation, not when the user types /issue directly.

{
  "hooks": {
    "PreToolUse": [
      {
        "matcher": "Skill",
        "hooks": [{ "type": "command", "command": "bash .claude/hooks/persona-issue-hook.sh" }]
      }
    ]
  }
}

Proposed Hook Contract

{
  "hooks": {
    "PreSkillUse": [
      {
        "matcher": "issue",
        "hooks": [{ "type": "command", "command": "bash .claude/hooks/persona-issue-hook.sh" }]
      }
    ]
  }
}

stdin payload (same pattern as PreToolUse):

{ "skill": "issue", "args": "Add swipe-to-add on ingredient categories" }

stdout from the hook is injected as context before the skill executes, allowing the hook to augment or block execution — same semantics as PreToolUse.