CodeScope v0.2.0 — Investigate LLM and VS Code Prototype

CodeScope v0.2.0 expands the retrieval-first debugging workflow with natural-language LLM investigation support and a VS Code extension prototype.

Highlights:

• Added investigate --llm
• Added investigate --json --llm
• Added investigation-specific LLM context and prompt builders
• Added shared redaction/context-safety helpers
• Added fake/OpenAI LLM support for both diagnose and investigate workflows
• Added VS Code TreeView extension prototype with workspace indexing, rebuild index, investigation results, and clickable file navigation
• Added empty-index recovery with index --rebuild
• Improved user-facing investigation reason text
• Updated README and docs for LLM investigation and extension behavior

Validation:

• 278 tests passing
• ruff passing
• benchmark evaluator: 3 PASS, 0 PARTIAL, 0 FAIL
• VS Code extension TypeScript compile passing
• External real-field testing on multiple Python projects

CodeScope remains retrieval-first: deterministic results are the source of truth, and optional LLM output explains retrieved/redacted context only. It does not generate patches or modify files.

CodeScope v0.1.0 — Retrieval-first Python debugging

CodeScope v0.1.0 is the first public release of a retrieval-first Python debugging tool.

Highlights:

• AST-based Python repository indexing
• Semantic code search
• Failure-aware pytest diagnosis
• Natural-language bug investigation
• JSON output for tools and future editor integrations
• Realistic benchmark evaluator
• Optional fake and OpenAI LLM providers
• Redacted/capped LLM context
• GitHub Actions CI

Validation:

• 247 tests passing
• ruff passing
• benchmark evaluator: 3 PASS, 0 PARTIAL, 0 FAIL

CodeScope does not generate patches or modify files. Optional LLM output is used only as reasoning over retrieved context.