claude -p returns empty result field despite model generating response

Found 3 possible duplicate issues:

1. https://github.com/anthropics/claude-code/issues/38769
2. https://github.com/anthropics/claude-code/issues/38725
3. https://github.com/anthropics/claude-code/issues/38760

This issue will be automatically closed as a duplicate in 3 days.

• If your issue is a duplicate, please close it and 👍 the existing issue instead
• To prevent auto-closure, add a comment or 👎 this comment

🤖 Generated with Claude Code

Reproducing on macOS (arm64) — additional environment details and isolation tests

Environment

• Claude Code: 2.1.83
• macOS 15 (Darwin 25.2.0), Apple Silicon, Mac Mini
• Auth: claude.ai OAuth (Max subscription)
• Last working version: 2.1.81
• Settings: effortLevel: "high", 9 plugins, 13 MCP servers, 3 hooks

Reproduction

Text mode — single newline byte on stdout:
claude -p "Say hello"

JSON mode — empty result, tokens consumed:

  claude -p "Say hello" --output-format json
  Returns: {"result": "", "output_tokens": 13, ...}

Stream-JSON — text IS generated, visible in assistant events:

  claude -p "Say hello" --output-format stream-json --verbose
  Assistant event: {"type":"assistant","message":{"content":[{"type":"text","text":"\n\nHello! How can I
   help you?"}]}}
  Final result event: {"result": ""}

Isolation tests (all empty on 2.1.83, all work on 2.1.81)

• Different models (Opus 4.6, Sonnet 4.6, Haiku 4.5) → Empty result
• Different effort levels (--effort low, --effort high) → Empty result
• Different directories (/tmp, ~/, project dirs) → Empty result
• No MCP (--strict-mcp-config --mcp-config '{"mcpServers":{}}') → Empty result
• No slash commands (--disable-slash-commands) → Empty result
• Explicit stdin (echo "" | claude -p "Say hello") → Empty result
• Same tests on v2.1.81 → All work correctly

Key observations

• output_tokens is always > 0 (model generates tokens)
• Content blocks are type: "text" (not thinking blocks — extended thinking is not the cause)
• stop_reason: "end_turn" — model completes normally
• is_error: false, subtype: "success" — CLI reports success
• Text exists in the streaming pipeline and is lost during final result assembly

Confirms this is not caused by:

• Extended thinking
• MCP servers or plugins
• Hooks (one hook shows Permission denied on session start, but disabling all hooks doesn't fix it)
• Model choice or effort level

Impact

Any programmatic use of claude -p that reads stdout or parses the JSON result field gets an empty
response. This breaks CI/CD pipelines, agent frameworks spawning claude -p as a subprocess, and
automation scripts.

Workaround

Pin to v2.1.81 under a different filename (the auto-updater overwrites both symlinks and binary copies

   at ~/.local/bin/claude):
  cp ~/.local/share/claude/versions/2.1.81 ~/.local/bin/claude-pinned

Then point your tooling at claude-pinned instead of claude.``

There are duplicated issues to this one - not sure which one will be the main one. Leaving this open as it already has thumb-up

This was fixed in v2.1.84 — Fixed claude -p returning empty result when Stop hooks (or other progress/attachment-emitting sources) are configured — result extraction now skips trailing progress/attachment messages to find the actual assistant response. If you're still seeing this in the latest version, please comment with your version and repro and we'll reopen.

This issue has been automatically locked since it was closed and has not had any activity for 7 days. If you're experiencing a similar issue, please file a new issue and reference this one if it's relevant.