Your AI skills are static. SLM changes that. Track which skills work, which fail, and watch them improve — session by session, automatically.
Zero-LLM, zero-cloud. Pure local statistical analysis that gets smarter every session.
Enriched hook captures every tool call with input, output, session context, and project path. Secret scrubbing built-in. Zero cost.
SkillPerformanceMiner builds execution traces, computes outcome heuristics, tracks per-skill metrics. Runs during consolidation — no latency impact.
Soft prompts route to high-performing skills. Behavioral assertions inform future sessions. Skill entities in Entity Explorer show the full picture.
The backend is IDE-agnostic. Any client can POST tool events. The shipped hook currently supports Claude Code.
| IDE | Status | Integration |
|---|---|---|
| Claude Code | Supported | Auto-registered via slm init |
| Any IDE | API Available | POST to /api/v3/tool-event |
| Cursor | Planned | Adapter in development |
| Windsurf | Planned | Adapter in development |
| VS Code / JetBrains | Planned | Extension adapter |
Everything Claude Code (ECC) is the most popular plugin for Claude Code with continuous learning, instinct-based pattern detection, and deep observation capabilities.
SLM integrates directly with ECC observations for richer skill tracking. One command imports all your ECC data into SLM's skill performance pipeline.
ECC is not required — SLM is fully self-sufficient. This integration is an optional enhancement for users who want both systems working together.
# Import ECC observations into SLM
$ slm ingest --source ecc
Ingested: 11,327 events from ecc
Files scanned: 15 projects
# Preview without writing
$ slm ingest --source ecc --dry-run
Would ingest: 11,327 events (dry run)
SLM's skill evolution architecture draws from cutting-edge academic research in self-evolving agent systems.
Co-evolutionary verification with information isolation. +30pp improvement from blind verification.
arXiv:2604.016873-trigger self-evolving skill engine. Anti-loop guards. Version DAG model. MIT license.
github.com/HKUDS/OpenSpace86-task benchmark: self-generated skills provide zero benefit without verification. Focused skills outperform.
arXiv:2602.12670Four-axis taxonomy of agentic skills. Skills and MCP are orthogonal layers. 26.1% vulnerability rate.
arXiv:2602.12430Install SLM. Your skills start learning from the very first session.
Open source under AGPL v3 — A Qualixar Research Initiative
