Skip to content

LeanCTL v0.2.0

Choose a tag to compare

View all tags
@yvgude yvgude released this 12 Apr 20:42
v0.2.0
960763f

lean-ctx v3.0.0 Integration — Full Premium Engine

This release fully integrates lean-ctx v3.0.0, bringing 8 new LLM-accessible tools,
automatic session intelligence, and quality-guarded compression to the agent loop.

New Tools (LLM-accessible)

  • impact_analysis — Graph-based impact analysis: "What breaks when file X changes?" Supports analyze, chain, build, and status actions via the property graph.
  • architecture — Project architecture analysis: module clusters, dependency layers, circular dependency detection, entrypoints.
  • file_heatmap — File access heatmap tracking which files are read most during a session. Helps the agent focus on hot paths.
  • cost_report — Token cost attribution across agents and tools. Shows where tokens are spent.
  • a2a_task — A2A (Agent-to-Agent) task management: create, assign, update, cancel, and track tasks between agents.
  • agent_coordination — Multi-agent coordination bus: register agents, post messages, handoff tasks, diary entries for agent work logs.
  • sandboxed_execute — Execute code in a sandboxed environment. Supports Python, JavaScript, Bash, Rust, and more.

Automatic Session Intelligence

  • Session findings — Tool errors are automatically recorded as session findings for cross-session debugging context.
  • Session decisions — File edits (write_file, edit_file, multi_edit) are automatically recorded as session decisions.
  • Turn summaries — Each turn's tool count, duration, and token savings are persisted as findings.
  • Checkpointing — Compressed context snapshots every 15 tool-loop iterations for recovery.
  • Task persistence — Current user task is recorded at the start of each turn via set_task.

Compression Pipeline

  • Quality guard — Structural integrity checks on compressed output. Rejects destructive compression and falls back to original.
  • Engine-compressed skip list — All lean-ctx tools skip redundant secondary compression, eliminating double-compression overhead.
  • Signatures mode — 94.2% token savings when reading files for API surface only.
  • Cache hits — 99.8% savings on re-reads (9 tokens vs ~3,864 average per file).

Under the Hood

  • Updated lean-ctx submodule from v1.8.0-318 to v3.0.0
  • Enabled embeddings feature (ONNX Neural Compression)
  • Added rusqlite for Property Graph support
  • Exported 5 new core modules: a2a, heatmap, property_graph, deep_queries, memory_lifecycle
  • Migrated all direct lean_ctx::core::* imports to engine wrapper layer
  • Removed dead engine/share.rs wrapper (no single-agent use case)
  • 72 tests passing (29 new v3 integration tests)

Install

curl -fsSL https://leanctl.com/install.sh | sh
npm install -g leanctl-bin
yay -S leanctl-bin  # Arch Linux

Full Changelog: v0.1.0...v0.2.0

Assets 7
Loading