wip: capture OpenViking repo state

[kal3w]

Preserves the current local OpenViking repo state in a dedicated capture branch.

[github-actions]

PR Reviewer Guide 🔍

Here are some key observations to aid the review process:

⏱️ Estimated effort to review: 4 🔵🔵🔵🔵⚪

🏅 Score: 78

🧪 No relevant tests

🔒 No security concerns identified

✅ No TODO sections

🔀 Multiple PR themes Sub-PR theme: MCP Bridge Enhancements and Codex Setup Relevant files: examples/mcp-query/server.py examples/common/recipe.py examples/mcp-query/README.md docs/en/guides/09-codex-setup.md docs/en/guides/06-mcp-integration.md Sub-PR theme: Claude Memory Plugin Offline Mode and Detached Commits Relevant files: examples/claude-memory-plugin/scripts/ov_memory.py examples/claude-memory-plugin/hooks/common.sh examples/claude-memory-plugin/hooks/session-start.sh examples/claude-memory-plugin/hooks/session-end.sh examples/claude-memory-plugin/hooks/stop.sh examples/claude-memory-plugin/hooks/user-prompt-submit.sh examples/claude-memory-plugin/hooks/hooks.json examples/claude-memory-plugin/README.md examples/claude-memory-plugin/.claude-plugin/marketplace.json Sub-PR theme: Add Timestamps to MatchedContext Relevant files: openviking_cli/retrieve/types.py openviking/retrieve/hierarchical_retriever.py

⚡ Recommended focus areas for review Type Mismatch Bug The _query_sync function returns a dict when recipe.query_ready is false, but the query MCP tool expects a string return type. This will cause a runtime error when the query tool is used without LLM config. if not recipe.query_ready: return { "answer": ( "Query is not configured on this MCP bridge. " "Provide a local ov.conf with vlm.api_base and vlm.model, " "or use the search tool and let Codex synthesize the answer." ), "context": [], "timings": {}, } N+1 Query Performance Issue The search method calls self.client.stat() for each search result to retrieve timestamps, leading to N additional roundtrips. This can degrade performance when many results are returned. Silent Exception Swallowing The _replay_offline_sessions function uses broad except Exception: clauses without logging, which hides errors during offline session replay and makes debugging difficult. except Exception: continue if not turns: turns_file.unlink(missing_ok=True) continue try: with OVClient(backend, ov_conf_path) as cli: sess = cli.create_session() sid = _as_text(sess.get("session_id")) if not sid: continue for t in turns: cli.add_message(sid, t["role"], t["content"]) cli.commit_session(sid) turns_file.unlink(missing_ok=True) replayed += 1 except Exception: break # server may have gone offline mid-replay Set -e Removed The set -e flag was removed from the common.sh script, which means the script will continue executing even if a command fails, potentially leading to unexpected behavior. set -uo pipefail

[github-actions]

PR Code Suggestions ✨

No code suggestions found for the PR.

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}