[FEATURE] Skill tool blocked in plan mode despite being read-only documentation loader

Resolved 💬 3 comments Opened Oct 21, 2025 by matthew-bram Closed Jan 8, 2026

Preflight Checklist

  • [x] I have searched existing requests and this feature hasn't been requested yet
  • [x] This is a single feature request (not multiple features)

Problem Statement

Title: Skill tool blocked in plan mode despite being read-only documentation loader

Description:

## Summary
The Skill tool is being blocked when plan mode is active, even though skills are designed to load documentation/guidance and don't modify system state.

## Steps to Reproduce

  1. Activate plan mode (user preference to review before execution)
  2. Attempt to invoke a skill using the Skill tool
  3. Observe: Tool use is rejected with "The user doesn't want to proceed with this tool use"

## Expected Behavior
Skills should be allowed in plan mode because they:

  • Load documentation from .claude/skills/{skill-name}/SKILL.md
  • Provide workflow guidance and structure
  • Don't modify any files or system state
  • Are essentially "read-only" operations

## Actual Behavior
Skill tool invocation is blocked by plan mode guard, treating it as a state-modifying operation.

## Impact

  • Users can't leverage skill-based workflows when planning
  • Feature-builder skill can't guide Phase 0 architecture planning
  • Skills designed for "proactive activation" can't trigger in plan mode

## Workaround
Exit plan mode, activate skill, then re-enter plan mode (or follow skill guidance manually)

## Environment

  • Claude Code CLI
  • Model: claude-sonnet-4-5-20250929
  • Skills tested: feature-builder (confirmed to exist and be properly configured)

Why This Matters

The feature-builder skill is specifically designed to:

  1. Prevent coding without architecture (Phase 0 gates)
  2. Guide planning before execution
  3. Auto-activate when feature work is detected

Blocking it in plan mode defeats its purpose - you want the skill active during planning to enforce Phase 0 architecture documentation before any code execution!

Good catch on this one - it's a genuine workflow blocker for the skill system. 👍

Proposed Solution

I'd like a flag or skill descriptor field that indicates the skill is for read-only use. I'm using the new skill I've created to help in feature design, which involves complicated analysis of our codebase without risk of code change until the proposed architecture changes have been approved.

Alternative Solutions

Use in edit mode, then switch to planning once the skill has been loaded into context.

Priority

Medium - Would be very helpful

Feature Category

CLI commands and flags

Use Case Example

_No response_

Additional Context

<img width="1382" height="630" alt="Image" src="https://github.com/user-attachments/assets/da5db9c0-25ed-4c48-a566-7e1ed5a9913b" />

View original on GitHub ↗

This issue has 3 comments on GitHub. Read the full discussion on GitHub ↗