OpenClaw provider returns empty after 2026-09-01 SQLite session storage migration

[systemize-ai]

Summary

OpenClaw migrated its session storage from per-agent JSONL files to a per-agent SQLite database on 2026-09-01 (release 2026.8.1 per meta.lastTouchedVersion, migration artifacts in ~/.openclaw/session-sqlite-migration-runs/). CodeBurn's OpenClaw provider still only reads the legacy path documented in the README:

OpenClaw | ~/.openclaw/agents/*.jsonl (legacy .clawdbot, .moltbot, .moldbot) | ...

Since the migration, OpenClaw no longer writes any .jsonl files under ~/.openclaw/agents/. The old ones were either moved to a session-sqlite-import-archive/ subfolder or renamed to *.jsonl.deleted.<timestamp>. New session data lives exclusively in ~/.openclaw/agents/<agent>/agent/openclaw-agent.sqlite (tables include transcript_events, session_transcript_active_events, session_conversations, etc.).

Result: codeburn doctor reports the openclaw provider as completely empty, even though the tool is actively used and has substantial real usage.

Steps to reproduce

1. Have OpenClaw >= 2026.8.1 installed and used, with sessions migrated to SQLite (any install where ~/.openclaw/agents/*/agent/openclaw-agent.sqlite exists and ~/.openclaw/agents/*.jsonl does not).
2. Run codeburn doctor --json --no-color.

Observed

{
  "provider": "openclaw",
  "displayName": "OpenClaw",
  "status": "empty",
  "probePaths": [],
  "candidatesFound": 0,
  "verdict": "NOTHING FOUND (tool likely not installed or no history yet)"
}

on codeburn 0.9.24 (also reproduced on 0.9.19). Confirmed the tool is installed and actively used (multiple agent SQLite DBs with recent mtime, GB of transcript data).

Expected

OpenClaw usage/model data continues to be discovered and reported after the storage migration.

Environment

• codeburn: 0.9.24 (also reproduced on 0.9.19)
• OpenClaw: 2026.8.1 (self-hosted "Mr. Krabs" build)
• OS: Ubuntu 24.04, Node 24.20.0

Notes / possible approach

• This is a different failure mode than doctor: extend probeRoots() beyond 7/42 providers to kill the silent-zero class #899 (which is about missing probeRoots() diagnostics for silent-zero providers, and lists openclaw as a Tier-2 candidate for that work). Even with probeRoots() added, the underlying JSONL glob would still resolve to zero files post-migration — the fix needs actual SQLite parsing support for the new storage, not just better diagnostics on the old path.
• Since several other providers already have SQLite parsers (Forge's ~/.forge/.forge.db, Zed's threads.db, Copilot's OTel store), there's precedent in the codebase for this shape of provider.
• Happy to share anonymized schema/sample rows from openclaw-agent.sqlite (transcript_events / session_transcript_active_events) if useful for scoping the parser.