Netclaw 0.5.0
0.5.0 2026-03-13
Netclaw v0.5.0 — Observer-driven memory recall, local model reliability, and skills platform overhaul
Memory Recall Planning
- Added sidecar-driven memory observation and recall planning with deterministic policy gates, expiry handling, and SQLite-backed retrieval — the agent now plans what to recall before acting, rather than issuing ad-hoc memory queries. (#198)
- Fixed memory curation worker shutdown:
TaskCanceledExceptionfrom in-flight store awaitables during daemon stop is now caught and discarded cleanly instead of surfacing as a crash log. (#190)
Local Model Reliability
- Fixed multi-turn tool calling for OpenAI-compatible providers: assistant messages with tool calls and tool result messages now serialize
tool_callsandtool_call_idcorrectly — previously the second LLM request in a tool loop contained malformed history causing models to fall back to emitting tool calls as raw text. (#210) - Added text-based tool call extraction for models that emit tool calls as XML-like prose instead of structured fields (observed with Qwen3.5) — the parser activates as a fallback when no structured tool calls are present, making these models usable for multi-step tool workflows. (#213)
- Added OpenAI-compatible local endpoint support — operators can now configure endpoints like Ollama or LM Studio as a first-class provider type. (#208)
- Fixed capability detection for GGUF and quantized model IDs: the normalizer now strips
.ggufextensions, GGML quantization suffixes (-Q5_K_M,-IQ2_XXS,-Q4_K_XL), and build-variant segments (-UD,-BPW4) before capability resolution — models now correctly report multimodal context windows and input types instead of falling back to text-only defaults. (#215)
Skills Platform
- Adopted the AgentSkills.io SKILL.md standard: skills now use YAML frontmatter (name, description, triggers, license, compatibility, allowed-tools) and a directory-based layout (
skill-name/SKILL.md) with progressive disclosure viareferences/,scripts/, andassets/subdirectories. Feed manifest supports per-file SHA-256 verification for multi-file skills. (#216) - Added
capability-referencesystem skill — the agent's single authoritative reference for all 16 built-in tools across 5 grant categories, CLI commands, scheduling syntax, MCP discovery patterns, and session context. Also updatesself-diagnosticsto v1.2.0 with a cross-reference. (#217) - Added
search-citationsystem skill — guides the agent on when to useweb_searchversus training data, requires source URLs for all specific factual claims, and covers local search, travel, and product search verticals with progressive disclosure reference files. (#219)
Session Self-Awareness
- The agent's session ID (
{channelId}/{threadTs}) is now injected into the per-turn dynamic context layers, allowing the agent to be self-referential about its own session in tool calls and memory operations. (#218)
Stability
- Fixed a stream materializer actor leak in the SignalR session binding: stream stage actors were previously created under the global
StreamSupervisor-0and accumulated over the daemon's lifetime. A new per-sessionSignalRSessionActorscopes the materializer to the session lifecycle, so all stream children are torn down automatically when the session ends. (#192)