Skill invocations re-executed after conversation compaction

Bug Description

After conversation compaction, Claude re-executes skills (like /gh-bug) that were already completed earlier in the session, creating duplicate issues/actions.

Steps to Reproduce

1. In a long conversation, invoke a skill like /gh-bug with arguments
2. Continue working on other tasks until the conversation compacts
3. After compaction, Claude sees the skill invocation in system-reminder context and re-executes it

Expected Behavior

Skills that were already executed before compaction should not be re-executed after compaction.

Actual Behavior

The skill invocation is preserved in a <system-reminder> block:

<system-reminder>
The following skills were invoked in this session. Continue to follow these guidelines:

### Skill: gh-bug
...
ARGUMENTS: [original arguments]
</system-reminder>

Combined with the continuation prompt "Continue with the last task that you were asked to work on", Claude interprets this as a pending task and re-executes the skill.

Root Cause Analysis

1. Skill invocations are preserved in system-reminder blocks with "Continue to follow these guidelines"
2. The compaction summary doesn't indicate the skill was already executed
3. The continuation prompt instructs to "continue with the last task"
4. Claude sees an apparent active instruction and executes it again

Suggested Fix

Either:

• Remove skill invocation context from system-reminders after execution
• Mark completed skill invocations as "completed" or "executed"
• Include in compaction summary that specific skills were already executed (e.g., "Bug #XX was created for issue Y")
• Change the system-reminder wording from "Continue to follow these guidelines" to something that indicates historical context

Environment

• Claude Code CLI
• Model: claude-opus-4-5-20251101
• Observed multiple times across compaction events

🤖 Generated with Claude Code