context-budgeter scans a repository, ignores common generated directories, estimates tokens cheaply, ranks files for a task query, and writes a Markdown context budget report.
The token estimate is intentionally cheap and deterministic: roughly one token per four characters.
python -m context_budgeter /path/to/repo --query "fix oauth login" --budget 8000
python -m context_budgeter /path/to/repo --query "fix oauth login" --output context-report.md --recommend-ignoreInstalled script:
context-budgeter /path/to/repo --query "billing retry bug" --budget 4000python -m unittest discover -s testsNo network calls are required by the test suite.