[FEATURE] Allow Plan Mode (Plan agent) to access project and user skills

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

When entering Plan Mode, Claude activates the built-in Plan agent to reason about what needs to be done before writing any code. However, built-in agents (Plan, Explorer, Verify) cannot access skills defined in .claude/skills or user-level skills. Only custom subagents in .claude/agents can explicitly list skills in their frontmatter.

This means that when the Plan agent generates a plan, it has no awareness of project-specific conventions, architectural standards, or business rules that the team has encoded in skills. The resulting plan may be technically valid but misaligned with the actual standards of the project.

Proposed Solution

There should be a way to declare which skills the built-in Plan agent should load, either via CLAUDE.md or a dedicated config block. For example:

built_in_agents:
  plan:
    skills:
      - .claude/skills/architecture.md
      - .claude/skills/coding-standards.md

Alternatively, a simpler opt-in could be a plan_skills key at the top level of CLAUDE.md. The interface should make it clear which skills are active when Plan Mode is triggered.

Alternative Solutions

Currently I work around this by duplicating the most critical skill content directly inside CLAUDE.md. This does not scale well when there are many skills, and it defeats the purpose of having modular, reusable skills. Another option is to create a custom subagent that mimics Plan behavior and explicitly lists skills, but this loses the native Plan Mode UX entirely.

Priority

High - Significant impact on productivity

Feature Category

Configuration and settings

Use Case Example

1. I have a project with a .claude/skills/architecture.md skill that defines layered architecture rules and naming conventions.
2. I enter Plan Mode and ask Claude to plan a new feature.
3. The Plan agent generates a plan that ignores the architecture conventions because it cannot read the skill.
4. I then have to manually correct the plan or paste the skill content into the prompt.
5. With this feature, the Plan agent would load the declared skills and generate a plan already aligned with the project standards.

Additional Context

This limitation is documented in the official Claude Code skills course (anthropic.skilljar.com), which explicitly states that built-in agents cannot access skills at all. The workaround of duplicating content in CLAUDE.md is fragile and not maintainable at scale. Skills are designed to encode reusable context, and Plan Mode is precisely the step where that context is most needed.