OpenClaw and Hermes are tightening the boundaries around always-on agents
The strongest new May 9 pattern is not one isolated bug; it is a cluster of boundary repairs around agents that live in chats, browsers, and long-running sessions. OpenClaw PR #79645 keeps transcript redaction centralized at append time, #79649 reduces stale Telegram reply ancestry confusing old replies for the active conversation, #79658 only allows local TXT/JSON/YAML media sends after validation, and #79562 targets Discord queue backpressure plus transcript/session-store read bottlenecks. Hermes PR #22280 hardens Telegram model-picker callback authorization, validates explicit Chrome DevTools Protocol override endpoints before connecting or discovering, and protects detailed health diagnostics behind API-key auth. Hermes PR #22261 fixes Gemini fallback failures when parallel tool responses are split across turns.
Always-on agents fail in the seams: an old reply looks current, a diagnostic endpoint leaks runtime state, a browser endpoint is too eager to connect, a local document send bypasses validation, or a queue silently accumulates work. These fixes show the ecosystem moving from “can the agent act?” toward “can it act without crossing the wrong boundary?”
- OpenClaw PR #79645 repairs centralized appendSessionTranscriptMessage redaction and adds regression tests for transcript redaction behavior
- OpenClaw PR #79649 includes real Telegram QA showing stale reply ancestry no longer dominates the active live chat window
- OpenClaw PR #79658 allows TXT/JSON/YAML local media sends only with MIME/extension validation and binary-disguise rejection coverage
- OpenClaw PR #79562 adds targeted backpressure and read-path fixes for Discord queues, media downloads, preview writes, transcript idempotency, Codex RPC writes, and session-store reads
- Hermes PR #22280 explicitly lists three security boundaries: Telegram callback authorization, safe CDP override validation, and auth for /health/detailed
- Hermes PR #22261 documents a Gemini HTTP 400 failure mode after parallel tool calls and coalesces tool responses into one user turn for both native and CloudCode adapters
- Most items are PR-stage at aggregation time and may not be in a tagged release yet
- The fixes span several surfaces, so teams should not cherry-pick blindly without matching their actual channels and providers
- Boundary hardening can expose previously hidden configuration mistakes, especially around API keys, browser automation endpoints, and chat history limits