[BUG] ${CLAUDE_EFFORT} in a skill preloaded into a subagent expands to the session effort, not the subagent's effort

Preflight Checklist

• [X] I have searched existing issues and this hasn't been reported yet
• [X] This is a single bug report (please file separate reports for different bugs)
• [X] I am using the latest version of Claude Code

What's Wrong?

${CLAUDE_EFFORT} in a skill preloaded into a subagent (via the agent's skills: frontmatter) expands to the parent session's effort, not the subagent's own effort. The subagent's effort is set by its effort: frontmatter and is what the hook payload reports (effort.level) for that subagent's tool calls, so the two disagree.

The docs describe ${CLAUDE_EFFORT} as the current effort level (https://code.claude.com/docs/en/skills) and document a subagent's own effort (https://code.claude.com/docs/en/sub-agents#choose-an-effort-level).

Not tested: a skill loaded through the Skill tool inside a subagent, and a spawn that passes effort.

What Should Happen?

The placeholder should expand to the effort the subagent is actually running at (low in the repro below).

Error Messages/Logs

Session effort: medium
Subagent frontmatter effort: low
Preloaded skill body: "Effort: ${CLAUDE_EFFORT}"  -> expanded to "Effort: medium"
Hook payload effort.level for the subagent's tool calls: "low"

Steps to Reproduce

1. Start a session at effort medium (/effort medium).
2. Create .claude/agents/effort-probe.md:

---
name: effort-probe
description: Reports the effort value its preloaded skill shows
effort: low
skills:
  - print-effort
---
Reply with the single line the print-effort skill gave you, nothing else.

1. Create .claude/skills/print-effort/SKILL.md:

---
name: print-effort
description: Prints the effort level
user-invocable: false
---
Effort: ${CLAUDE_EFFORT}

1. Prompt: Use the effort-probe agent and show me its reply.
2. The reply is Effort: medium. A PreToolUse/PostToolUse hook on the subagent's tool calls receives effort.level = low.

Automated version, from the public repo melodic-software/claude-code-plugins (on the branch of PR #6913, or main once it lands):
python3 plugins/harness-ops/skills/behavior-probes/scripts/probe.py run --live --case subagent/preloaded-skill-placeholder-reflects-session-effort

Claude Model

Not sure / Multiple models

Is this a regression?

I don't know

Claude Code Version

2.1.296 (Claude Code)

Platform

Anthropic API

Operating System

Other Linux

Terminal/Shell

Other