Subagents should inherit parent session hooks and permission rules

[thomascburns]

Problem

When using the Task tool to launch subagents (Explore, Plan, general-purpose, Bash), the subagents do not inherit:

1. PreToolUse hooks defined in .claude/settings.json
2. Permission allow/deny rules from .claude/settings.json
3. CLAUDE.md instructions
4. Project memory files

This means project-level safety enforcement (e.g., a hook that blocks cd to the project root, or deny rules for destructive commands) silently stops working when subagents are involved.

Current Behavior

• Parent session has a PreToolUse hook on Bash that blocks cd to the project root
• Parent session has permission rules allowing read-only commands without prompting
• Subagents launched via Task(subagent_type="Explore") ignore both — they cd into the project root freely and prompt for permission on commands the parent auto-allows

Expected Behavior

Subagents should inherit all hooks and permission rules from the parent session's .claude/settings.json by default, ensuring consistent enforcement across the entire agent tree.

Current Workaround

Creating custom agent definitions in .claude/agents/ that manually replicate each hook:

---
name: explore
hooks:
  PreToolUse:
    - matcher: "Bash"
      hooks:
        - type: command
          command: "\"$CLAUDE_PROJECT_DIR\"/.claude/hooks/no-cd-project-root.sh"
---

This is fragile because:

• Every new hook in settings.json must be manually duplicated into every agent file
• The agent definitions can drift out of sync with the actual settings
• New subagent types require new override files
• It embeds project rules in prose that the AI may not follow, whereas hooks are deterministic

Proposed Solution

Add a configuration option (or make it the default) to propagate parent session hooks and permission rules to all subagents:

{
  "permissions": {
    "propagateToSubagents": true
  },
  "hooks": {
    "propagateToSubagents": true
  }
}

Or simply make inheritance the default behavior, since hooks and permissions represent project-level safety constraints that should apply everywhere.

Environment

• Claude Code CLI
• macOS (Darwin 25.3.0)
• Using .claude/settings.json with PreToolUse hooks and permission allow/deny rules

[github-actions]

Found 3 possible duplicate issues:

1. [SECURITY] PreToolUse hooks not enforced on subagent tool calls, allowing security bypass #21460
2. Skills/subagents do not inherit user-level permissions from settings.json #18950
3. [FEATURE] --freeze-context / --context-lock for subagent and agent team configuration inheritance #27533

This issue will be automatically closed as a duplicate in 3 days.

• If your issue is a duplicate, please close it and 👍 the existing issue instead
• To prevent auto-closure, add a comment or 👎 this comment

🤖 Generated with Claude Code

[thomascburns]

Closing as duplicate. This is covered by the combination of:

• [SECURITY] PreToolUse hooks not enforced on subagent tool calls, allowing security bypass #21460 — PreToolUse hooks not enforced on subagent tool calls
• Skills/subagents do not inherit user-level permissions from settings.json #18950 — Skills/subagents do not inherit permissions from settings.json
• [FEATURE] --freeze-context / --context-lock for subagent and agent team configuration inheritance #27533 — --freeze-context for subagent configuration inheritance

Upvoted all three. The workaround (custom .claude/agents/*.md with hooks replicated in frontmatter) is documented in our project.

[github-actions]

This issue has been automatically locked since it was closed and has not had any activity for 7 days. If you're experiencing a similar issue, please file a new issue and reference this one if it's relevant.