feat(conversation): Add /condense command as a skill-triggered CondensationRequest

[enyst]

Summary

• Implement explicit /condense user command handling in LocalConversation.send_message
• Leverage skills system: activation of a skill named "condense" also triggers condensation
• Emit CondensationRequest event and return early (no user MessageEvent) to let condenser act next run
• Add comprehensive unit tests for the flow
• Provide runnable example using Gemini 2.5 Pro condenser

Changes

1. LocalConversation

• Detect "/condense" in user message text content OR activation of a knowledge skill named "condense"
• When detected, emit CondensationRequest() and return, leaving the next run() to process condensation via configured condenser

2. Tests

• tests/sdk/conversation/local/test_condense_command.py
  • test_send_message_with_slash_condense_emits_request: ensures CondensationRequest is emitted and no MessageEvent is recorded for that call
  • test_condense_request_triggers_condenser_on_next_step: with LLMSummarizingCondenser configured, verifies a Condensation event appears after run()
  • test_condense_skill_trigger_name: activating a skill named "condense" also emits CondensationRequest

3. Example

• examples/01_standalone_sdk/15_condense_command.py
  • Demonstrates explicit /condense command and skill-triggered condensation
  • Uses model gemini-2.5-pro with GEMINI_API_KEY from environment

Motivation
This PR implements the user-facing /condense command requested in issue #8276, aligning with the existing condenser flow (LLMSummarizingCondenser) and skills architecture. It allows users to explicitly request context condensation, especially useful with an LLM summarizer condenser.

How it works

• On send_message(), we first allow per-turn skill activation (to collect activated_skill_names)
• If a skill named "condense" is activated or the message includes "/condense", we emit CondensationRequest and exit
• On the next run(), the condenser notices the unhandled request and performs condensation, optionally injecting a CondensationSummaryEvent into the view

Testing

• Added focused tests; they pass locally:
  • uv run pytest tests/sdk/conversation/local/test_condense_command.py
• Ran pre-commit hooks on changed files (ruff format/lint, pyright): passed

Docs/Examples

• Included a standalone example using Gemini 2.5 Pro. To run:
  GEMINI_API_KEY=... uv run python examples/01_standalone_sdk/15_condense_command.py

Notes

• This change is backward-compatible; only adds a conditional shortcut for /condense
• The example uses gemini-2.5-pro; ensure appropriate API key is set in env

Closes #8276

@enyst can click here to continue refining the PR

---

Agent Server images for this PR

• GHCR package: https://github.com/OpenHands/agent-sdk/pkgs/container/agent-server

Variants & Base Images

Variant	Architectures	Base Image	Docs / Tags
java	amd64, arm64	eclipse-temurin:17-jdk	Link
python	amd64, arm64	nikolaik/python-nodejs:python3.12-nodejs22	Link
golang	amd64, arm64	golang:1.21-bookworm	Link

Pull (multi-arch manifest)

# Each variant is a multi-arch manifest supporting both amd64 and arm64
docker pull ghcr.io/openhands/agent-server:1f398f0-python

Run

docker run -it --rm \
  -p 8000:8000 \
  --name agent-server-1f398f0-python \
  ghcr.io/openhands/agent-server:1f398f0-python

All tags pushed for this build

ghcr.io/openhands/agent-server:1f398f0-golang-amd64
ghcr.io/openhands/agent-server:1f398f0-golang_tag_1.21-bookworm-amd64
ghcr.io/openhands/agent-server:1f398f0-golang-arm64
ghcr.io/openhands/agent-server:1f398f0-golang_tag_1.21-bookworm-arm64
ghcr.io/openhands/agent-server:1f398f0-java-amd64
ghcr.io/openhands/agent-server:1f398f0-eclipse-temurin_tag_17-jdk-amd64
ghcr.io/openhands/agent-server:1f398f0-java-arm64
ghcr.io/openhands/agent-server:1f398f0-eclipse-temurin_tag_17-jdk-arm64
ghcr.io/openhands/agent-server:1f398f0-python-amd64
ghcr.io/openhands/agent-server:1f398f0-nikolaik_s_python-nodejs_tag_python3.12-nodejs22-amd64
ghcr.io/openhands/agent-server:1f398f0-python-arm64
ghcr.io/openhands/agent-server:1f398f0-nikolaik_s_python-nodejs_tag_python3.12-nodejs22-arm64
ghcr.io/openhands/agent-server:1f398f0-golang
ghcr.io/openhands/agent-server:1f398f0-java
ghcr.io/openhands/agent-server:1f398f0-python

About Multi-Architecture Support

• Each variant tag (e.g., 1f398f0-python) is a multi-arch manifest supporting both amd64 and arm64
• Docker automatically pulls the correct architecture for your platform
• Individual architecture tags (e.g., 1f398f0-python-amd64) are also available if needed

[github-actions]

Coverage Report •

File	Stmts	Miss	Cover	Missing
openhands-sdk/openhands/sdk/conversation/impl
local_conversation.py	189	53	71%	121, 123–124, 157–158, 181, 186, 198, 215, 223–225, 229–230, 238–239, 291–292, 295, 302, 323–325, 328, 337, 353, 355, 357, 361, 363–365, 367, 369, 375–376, 389–390, 392, 394, 398–401, 419, 424, 426, 431, 433–435, 453, 455
TOTAL	11805	5430	54%

[openhands-ai]

Looks like there are a few issues preventing this PR from being merged!

• GitHub Actions are failing:
  • Check Documented Examples

If you'd like me to help, just leave a comment, like

@OpenHands please fix the failing actions on PR #1114 at branch `openhands/condense-command-skill`

Feel free to include any additional details that might help me get this PR into a better state.

_{^{You can manage your notification settings}}