release changelog discord security

OpenClaw v2026.2.15: Discord Components V2, Nested Sub-Agents & More

OpenClaws.io Team

OpenClaws.io Team

@openclaws

February 16, 2026

6 min read

OpenClaw v2026.2.15: Discord Components V2, Nested Sub-Agents & More

OpenClaw v2026.2.15 was released on February 16, 2026. This release introduces Discord Components V2 for native rich interactions, nested sub-agents with configurable spawn depth, LLM hook payloads for extension developers, and a broad set of security fixes. Over 40 contributors made this one happen.

Discord Components V2

Discord's Components V2 API is now fully supported. Buttons, select menus, modals, and attachment-backed file blocks let your agent deliver native interactive prompts without workarounds. CV2 containers, improved button layouts, and a refined exec-approval UX make the experience seamless. Thanks @thewilloftheshadow for driving this effort.

Nested Sub-Agents

Sub-agents can now spawn their own children. Set `agents.defaults.subagents.maxSpawnDepth: 2` to control how deep the chain goes. A `maxChildrenPerAgent` limit (default 5) prevents runaway spawning, depth-aware tool policies keep permissions tight, and announce-chain routing ensures messages reach the right parent. Thanks @tyler6204.

LLM Hook Payloads

Extension authors can now observe what goes into and comes out of the model. The new `llm_input` and `llm_output` hook payloads expose prompt context and model responses, opening the door to logging, guardrails, and analytics plugins. Thanks @SecondThread.

Security Improvements

This release includes significant security hardening across the stack:

  • SHA-256 sandbox hashing replaces deprecated SHA-1 for sandbox config integrity checks.
  • Telegram bot token redaction — tokens are now stripped from error messages and stack traces.
  • Dangerous Docker sandbox configs blocked — bind mounts, host networking, and unconfined seccomp/apparmor profiles are rejected.
  • Session/path redaction — sensitive details removed from status responses for non-admin clients.
  • LINE fails closed when channel token or secret is missing.
  • Skill installer sandboxed — downloads restricted to the per-skill tools directory.
  • Web fetch body cap — response body size is capped before HTML parsing to prevent memory abuse.
  • Stored XSS prevention via assistant name and avatar sanitization.
  • Workspace path sanitization before LLM prompt embedding.
  • Git pre-commit hook hardened against option injection from malicious filenames.

Platform Fixes

  • Telegram: fixed DM thread errors, duplicate messages, streaming glitches, and voice message handling.
  • Discord: session continuity improvements, deduplicated slash commands, role allowlist matching fixes.
  • Memory: Unicode-aware FTS queries with CJK support, timezone-aware memory dates.
  • Group chats: context now injected on every turn, not just the first.

Other Highlights

  • Per-channel ack reaction overrides for Slack, Discord, and Telegram.
  • Cron webhook delivery toggle and auth token support.
  • Browser tool returns explicit non-retry guidance when unavailable.
  • Full contributor list available on the GitHub release page.

Stay in the Loop

Get updates on new features, integrations, and lobster wisdom. No spam, unsubscribe anytime.