[BUG] Claude stops after skill output completion despite pending TODO items

Description

Claude Code stops execution and waits for user input after completing skill invocations, even when TODO items clearly indicate more work remains. This appears to be a variant of the behavior described in #1066 and #1632 (both now locked), specifically triggered by skill output completion.

Environment

• Claude Code version: 1.0.x
• Model: Opus 4.5
• OS: macOS
• Plugin: Custom plugin with skills that invoke sub-skills

Reproduction Steps

1. Create a plugin with a parent skill that invokes sub-skills (e.g., /project-status → /local-repo-check → /github-repo-check)
2. Start a session with a TODO list containing multiple steps
3. Ask Claude to execute the parent skill workflow

Expected Behavior

Claude should:

1. Invoke /local-repo-check, output results
2. Immediately invoke /github-repo-check, output results
3. Synthesize combined results
4. Continue to next TODO item

Actual Behavior

Claude stops after each skill outputs its markdown table:

1. /project-status invokes /local-repo-check
2. local-repo-check completes, outputs markdown table
3. Claude stops - waits for user input
4. User says "continue" or "why did you stop?"
5. /project-status then invokes /github-repo-check
6. github-repo-check completes, outputs markdown table
7. Claude stops again

TODO state at each stop: 4-6 items still marked pending

Frequency

Happened 4+ times across two separate sessions. Each skill output completion acted as an implicit stop point.

Hypothesis

The model interprets "output complete" (markdown table rendered) as a natural stopping point, emitting a stop token even though:

• The TODO list shows pending items
• The workflow has clear continuation steps
• No explicit instruction to pause was given

This is consistent with the stop token hypothesis in #1632 (@csullivan's analysis), but specifically triggered by skill output completion rather than arbitrary points in execution.

Related Issues

• #1066 - Claude stops and waits for user input with TODO list (LOCKED)
• #1632 - Claude stopping with unfinished TODOs (LOCKED)

Searched for open issues but found no current tracker for this behavior.

---

Reference

• Plugin: https://github.com/my-entourage/entourage-plugin/tree/35d7f402c150573911706ef3c4cd9f36a6775d91
• Skills involved:
• skills/project-status/SKILL.md
• skills/local-repo-check/SKILL.md
• skills/github-repo-check/SKILL.md
• Evaluation files:
• skills/project-status/evaluations/evaluation.json