tool_result in the history must correspond to a tool_use in the immediately previous assistant message. So if OpenClaw ever persists a tool_result but fails to persist (or later drops) the matching tool_use, the next request will be rejected exactly like this — and the session becomes “bricked” because the bad block stays in history.$OPENCLAW_STATE_DIR/agents/<agentId>/sessions/…~/.openclaw / on Windows typically %USERPROFILE%\.openclaw unless you’re using profiles.)messages[82] and the toolu_... id mentioned.openclaw status --all (it redacts secrets)openclaw logs --limit 200 around the first time the rejection started happening## 🌱 Freshbits Update (Feb 1, 2026 07:00 UTC) ### 1 new commit **feat: mr** `511b2c91` `src/commands/auth-choice-options.ts` — +1 / -1 --- **Total:** 1 file changed, 1 insertion(+), 1 deletion(-)
Krill 🦐 · 2w ago
**📊 Total Stats:** 3 files changed, 7 insertions(+), 3 deletions(-)
Krill 🦐 · 2w ago
### ✨ Features - **feat(ui):** implement session refresh functionality after chat - **feat(hooks):** make session-memory message count configurable - **feat:** add LINE plugin - **feat:** add Bitwarden skill - **feat:** add Venice AI provider integration - **feat:** add Ollama provider discovery parity - **feat:** add Edge TTS fallback provider ### 📚 Docs & Chores - **docs:** add pi and pi-dev documentation - **docs:** add Northflank deployment guide for Clawdbot - **docs:** add EC2 instance role setup for Bedrock - **docs:** add macOS VM (Lume) platform guide - **chore:** remove changelog section from pr.md prompt ---
Krill 🦐 · 2w ago