Skip to content

CHANGELOG.md

Latest commit

 

History

History
78 lines (62 loc) · 4.09 KB

CHANGELOG.md

File metadata and controls

78 lines (62 loc) · 4.09 KB

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

v0.2.0 — V2: Grounding & Multi-Context Workspaces

Added

  • Grounding (anti-hallucination): Verify code claims have graph backing before an agent asserts them.
    • codebase_ground: Verify a claim about code against the graph. Returns verified/partial/ungrounded with evidence.
    • codebase_evidence: Find graph evidence for a symbol name.
    • codebase_suggest: Suggest similar symbols for typos/hallucinations (Levenshtein distance).
  • Multi-context workspaces: Load and query multiple codebases simultaneously.
    • workspace_create: Create a multi-codebase workspace.
    • workspace_add: Add a codebase with role (source/target/reference/comparison).
    • workspace_list: List all loaded codebases.
    • workspace_query: Search across all graphs.
    • workspace_compare: Compare a symbol between source and target.
    • workspace_xref: Cross-reference where a symbol exists/doesn't.
  • Translation mapping: Track code migration progress (source->target).
    • translation_record: Record a source->target symbol mapping.
    • translation_progress: Get migration progress statistics.
    • translation_remaining: List symbols not yet ported.
  • 69 new V2 stress tests (grounding, workspace, translation, MCP integration).

Changed

  • MCP tool count increased from 5 to 17.

[0.1.5] - 2026-02-23

Fixed

  • Enforced strict MCP parameter validation for symbol_lookup.mode and impact_analysis.max_depth to prevent silent fallbacks.
  • Switched per-project graph identity to canonical-path hashing to eliminate graph collisions for same-named folders.
  • Removed unsafe cached-graph fallback that could bind the wrong project graph in multi-project sessions.
  • Added runtime compile locking in agentic-codebase-mcp and hardened launcher lock acquisition for concurrent startup reliability.
  • Added regression tests for deterministic/unique project identity keys.

[0.1.4] - 2026-02-23

Fixed

  • Hardened MCP graph lock handling to recover from stale lockfiles and avoid deadlock under concurrent launches.
  • Ensured repo graph resolution falls back safely when common root detection does not yield a graph path.
  • Improved per-repo auto-indexing reliability so graph_stats no longer fails with empty graph state during normal startup races.

[0.1.2] - 2026-02-23

Fixed

  • MCP list_units now enforces and validates unit_type filters consistently.
  • MCP impact_analysis now includes full dependency coverage across containment/semantic edges.
  • Added regression tests to lock both MCP fixes for future releases.

[0.1.1] - 2026-02-22

Fixed

  • Hardened MCP stdio framing to correctly handle Content-Length protocol messages.
  • Improved interoperability with desktop clients that send framed MCP requests.

Changed

  • Documentation updates for workspace orchestration and install profiles.

[0.1.0] - 2026-02-19

Added

  • Semantic code compiler with tree-sitter parsing for Python, Rust, TypeScript, and Go
  • 13 code unit types and 18 edge types for typed concept graphs
  • Binary file format (.acb) with 128-byte header, fixed-size records, LZ4-compressed string pools
  • Memory-mapped file access via memmap2
  • Query engine with 24 query types across three tiers (Core, Built, Novel)
  • Five index types: SymbolIndex, TypeIndex, LanguageIndex, PathIndex, EmbeddingIndex
  • Semantic analysis: cross-language resolution, pattern detection, visibility inference, FFI tracing
  • Temporal analysis: change history, stability scoring, coupling detection, failure prophecy
  • Collective intelligence: delta compression, pattern extraction, privacy filtering
  • CLI tool (acb) with compile, info, query, and get commands
  • MCP server (agentic-codebase-mcp) with JSON-RPC 2.0 over stdio
  • 386 tests (38 unit + 348 integration), 21 Criterion benchmarks
  • Research paper: "AgenticCodebase: A Semantic Code Compiler for Navigable, Predictive, and Collective Code Intelligence"