v0.1.3 — Structured Context Engine

This release replaces naive source concatenation with an AST-driven structured summary pipeline. Context quality is significantly higher while token usage is 4-5x lower.

What changed

Structured summaries for core modules Tier 1 files no longer dump raw source into CONTEXT.md. Each file now emits a deterministic, AST-derived summary covering purpose, internal dependencies, public types, function signatures, and behavioral notes. Entry point files (cli.py, main.py, app.py, etc.) retain full source since agents benefit from seeing execution wiring directly.

Smarter ranking Replaced static file-count caps (MAX_CORE_MODULES = 10) with budget-driven inclusion. Tier thresholds are now percentile-based rather than hardcoded score cutoffs, producing consistent tier distribution across repos of any size. Default scoring weights rebalanced — git frequency and fan-in each at 0.40, recency reduced to 0.10.

Cleaner output structure Tests, docs, and example directories are forced to Tier 3 (one-liner summaries) regardless of score. A new RANKED_FILES section in every CONTEXT.md shows each file's score, tier, and token cost — making the ranker's decisions fully transparent and auditable.

Bug fixes

• Fixed SYMBOL_INDEX corruption caused by malformed tree-sitter byte offset extraction
• Fixed double budget consumption where the formatter was re-charging the budget for already-compressed content
• Fixed _auto_architecture hardcoded "python-based" string — now detects dominant language from the file set

Code quality Resolved all ruff lint errors across cli.py, compressor/tiered.py, graph/resolver.py, ranker/scorer.py, and ranker/semantic.py.

---

Numbers

Running codectx analyze . on the codectx repo itself:

Metric	Before	After
Context tokens	~40,000	~8,000
Analysis time	—	11s
Files in periphery	mixed	tests + docs only

---

Upgrade

pip install --upgrade codectx
# or
uv add codectx

codectx v0.1.2 — CI stability and testing improvements

Release Notes

This release focuses on improving the reliability and reproducibility of the development and testing workflow. No functional changes have been introduced to the core functionality of codectx. The updates primarily ensure that the project’s continuous integration environment correctly mirrors the expected runtime environment.

Improvements

• CI dependency installation fixed
  The GitHub Actions workflow now installs the semantic optional dependency group during testing. This ensures that tests relying on lancedb and sentence-transformers run correctly in CI environments.

• Improved test environment consistency
  CI now more accurately reflects local development environments, reducing discrepancies between local test results and CI outcomes.

• Workflow reliability improvements
  The CI pipeline is now more robust and deterministic, ensuring that formatting, type checking, dead code detection, tests, and build verification execute consistently across supported Python versions.

• Code formatting automation
  The CI workflow automatically applies Ruff formatting and commits formatting fixes to pull request branches when needed.

Developer Experience

These changes improve the contributor experience by:

• Ensuring pull requests run against a fully configured environment
• Preventing false test failures caused by missing optional dependencies
• Maintaining consistent formatting across the codebase
• Providing clearer CI feedback for contributors

Supported Python Versions

codectx continues to support:

Python 3.10
Python 3.11
Python 3.12

Installation

Install the latest version using:

pip install codectx

or

uv add codectx