feat(manager): migrate Manager runtime from copaw to qwenpaw 2.0 - #1095
Open
LUOSENGWA wants to merge 9 commits into
Open
feat(manager): migrate Manager runtime from copaw to qwenpaw 2.0#1095LUOSENGWA wants to merge 9 commits into
LUOSENGWA wants to merge 9 commits into
Conversation
|
Caution The consumer version of Gemini Code Assist on GitHub has been sunset. All code review activity has officially ceased. |
LUOSENGWA
force-pushed
the
manager-qwenpaw-2.0
branch
from
July 28, 2026 14:16
aa3c80e to
0676a69
Compare
Manager now runs on QwenPaw 2.0.1 alongside PR agentscope-ai#1077's Worker migration. Key changes: - Separate venvs: /opt/copaw-venv (copaw 1.0.2 for bridge.py) and /opt/venv/qwenpaw (qwenpaw 2.0.1 for Manager runtime), avoiding agentscope version conflict (1.0.20 vs 2.0.4) - Replace copaw hooks (monkey-patch CoPawAgent._create_toolkit) with agentteams-manager-tools QwenPaw plugin (api.register_tool()) for projectflow/taskflow/message/filesync - Delete Matrix channel physical overlay; use QwenPaw plugin system (agentteams-matrix-channel) symlinked into WORKING_DIR/plugins/ - run_copaw_app.py simplified to runpy.run_module('qwenpaw') - bridge.py invoked with /opt/copaw-venv/bin/python3 (pure stdlib, copaw 1.0.2 deps preserved) - CI pgrep pattern accepts both 'copaw app' and 'qwenpaw app'
LUOSENGWA
force-pushed
the
manager-qwenpaw-2.0
branch
from
July 28, 2026 14:26
0676a69 to
9e293b2
Compare
The message and taskflow tools deferred-imported copaw.config.config to read Matrix credentials (homeserver, access_token, user_id). In the QwenPaw 2.0 venv copaw 1.0.2 is absent, so _matrix_config_for_agent() raised ImportError — the message tool could not send any Matrix messages. Fix: read workspaces/<id>/agent.json directly (framework-agnostic). bridge.py already writes the Matrix channel config there. Also add QWENPAW_WORKING_DIR fallback to _resolve_copaw_working_dir() and _current_actor().
copaw is the predecessor of qwenpaw (renamed package). In the qwenpaw 2.0 venv the copaw package does not exist, so the try/except in _resolve_copaw_working_dir() should attempt qwenpaw first.
Systematic alignment with PR agentscope-ai#1077 Worker Dockerfile and QwenPaw 2.0 conventions: Dockerfile: - CMS observability: loongsuite-instrumentation-copaw → -qwenpaw + add loongsuite-otel-util-genai (matches Worker Dockerfile) start-copaw-manager.sh: - Add QWENPAW_SECRET_DIR, QWENPAW_RUNNING_IN_CONTAINER, QWENPAW_LOG_LEVEL env vars (matches Worker entrypoint) - CMS bootstrap: replace python3 heredoc script with direct heredoc (matches Worker), add LOONGSUITE_PYTHON_SITE_BOOTSTRAP_LOG_SUCCESS export - Inject session file privacy policy into AGENTS.md/SOUL.md (matches Worker _ensure_session_file_prompt_policy) - Remove dead PYTHONPATH=/opt/agentteams/copaw/src (path never existed in the image; modules are in site-packages) agent.manager.json: - Move approval_level into running section (matches QwenPaw 2.0 AgentProfileConfig schema, default AUTO)
QA Agent: Worker disables QwenPaw_QA_Agent_0.2 via API client.
Manager runs QwenPaw in-process with no API client, so set
enabled=false in config.json agents.profiles before startup.
QwenPaw's ensure_qa_agent_exists() skips agents already in profiles,
so our preset is preserved. start_all_configured_agents() skips
enabled=false agents.
approval_level: Move from running{} to top-level in agent.manager.json.
QwenPaw 2.0 AgentProfileConfig.approval_level is a top-level field,
not inside AgentsRunningConfig (which has extra='ignore' and would
silently drop it).
OpenClaw Manager sets tools.exec.ask=off for YOLO mode. QwenPaw 2.0's equivalent is approval_level=OFF (short-circuits the governance pipeline to ALLOW-all). start-manager-agent.sh promotes the yolo-mode marker file to AGENTTEAMS_YOLO=1 before calling this script; we translate that to agent.json approval_level=OFF here.
After PR agentscope-ai#1095 the Manager runs QwenPaw 2.0, not copaw. The script name should reflect reality. start-manager-agent.sh now accepts both 'copaw' (legacy runtime identifier used by controller/installer/CI) and 'qwenpaw' (forward- looking name). Both map to the same start-qwenpaw-manager.sh. Controller, installer, CI, Makefile, and Docker image name still use 'copaw' as the runtime identifier — changing those is a separate PR (the controller already defines RuntimeQwenPaw='qwenpaw' for Workers; adding a qwenpaw Manager runtime requires new config fields, image selection logic, and CI matrix entries).
…disable Six issues found during deep audit with CI script and PR agentscope-ai#1077 cross-reference: 1. pgrep pattern: \|qwenpaw app was a literal pipe in ERE, breaking the original copaw pattern. Manager process cmdline is still 'python3 -m copaw_worker.run_copaw_app app' (runpy.run_module does not change /proc/cmdline), so the original agentscope-ai#1077 pattern works. Reverted to 'copaw(_worker\.run_copaw_app)? app'. 2. Plugins invisible at runtime: Dockerfile copied plugins to /root/manager-workspace/.copaw/plugins/ (build-time), but the install script mounts a host volume at /root/manager-workspace, hiding build-time files. Now stores plugins in image-local /opt/agentteams/plugins/ and copies them at runtime in start-qwenpaw-manager.sh (mirrors Worker's _prepare_builtin_plugin). 3. CMS packages in wrong venv: pip install ran while PATH pointed to copaw-venv, but Manager runs from qwenpaw venv. Moved CMS install after qwenpaw venv creation with /opt/venv/qwenpaw/bin/pip. 4. QA Agent disable silently fails: jq wrote {"enabled": false} but AgentProfileRef requires id + workspace_dir. Pydantic validation stripped the incomplete profile, then ensure_qa_agent_exists() recreated it with enabled=true. Now writes a complete profile. 5. taskflow._current_actor() working dir resolution inconsistent with message.py: only checked env vars, no qwenpaw.constant fallback. Added _resolve_working_dir() mirroring message.py's logic. 6. plugin.py docstring stale: claimed taskflow/message still import copaw.config.config, but commit 974cdf6 already removed that. plugin.json type changed from 'general' to 'tool'.
…migration overwrite The QA Agent disable jq only injected agents.profiles['QwenPaw_QA_Agent_0.2'] without the 'default' profile. On first boot, QwenPaw's migrate_legacy_workspace_to_default_agent() sees len(profiles)==1 and 'default' not in profiles → runs legacy migration → replaces config.agents entirely → QA profile lost → ensure_qa_agent_exists() re-creates it with enabled=True, defeating the disable. Fix: inject both 'default' and QA profiles so len(profiles)>=2, which causes the migration to be skipped (multi-agent config detected).
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Migrate the Manager container from copaw 1.0.2 to QwenPaw 2.0.1, following the same plugin-based architecture that PR #1077 established for Workers.
What changed
Manager runtime — separate venv for QwenPaw 2.0.1 to avoid agentscope version conflicts with copaw 1.0.2 (used only by bridge.py for config conversion). Manager tools (projectflow, taskflow, message, filesync) registered via a QwenPaw plugin instead of monkey-patching
CoPawAgent._create_toolkit.Matrix channel — physical overlay replaced with the
agentteams-matrix-channelplugin (same as Workers). CMS observability packages installed in the QwenPaw venv (aligned with Worker).Configuration —
QWENPAW_WORKING_DIRset alongside legacyCOPAW_WORKING_DIR. Session-file privacy policy injected into prompts.approval_level=AUTOat top level of agent template. YOLO mode bridged toapproval_level=OFF. Built-in QA Agent disabled. DeadPYTHONPATHentries removed.Tool compatibility —
messageandtaskflowtools read Matrix credentials directly fromagent.jsoninstead of importingcopaw.config.config(not available in qwenpaw venv). Working directory resolution handles bothQWENPAW_WORKING_DIRandqwenpaw.constantfallback.CI — pgrep health check remains
copaw(_worker\.run_copaw_app)? app(Manager process cmdline is unchanged).Plugins at runtime — stored image-local and copied to working dir on startup, since the workspace is a host-mounted volume.
Stack
Based on
codex/qwenpaw-2-startup(PR #1077). Will retarget tomainafter #1077 merges.Follow-up
Controller/install/Makefile runtime identifier migration (
copaw→qwenpawfor Manager) is tracked for a separate PR — it touches Go controller code that #1077 also modifies, and should follow the same pattern (addqwenpawalongsidecopaw, then deprecate).