v0.1.0

FastContext now scopes exploration to an explicit repository and returns source-backed citations instead of unchecked model output. Idle MCP connections no longer retain model weights.

Changes:

• Add work_dir for repository/worktree selection, repository-scoped Read/Glob/Grep tools, and ripgrep discovery from PATH.
• Validate citations against source lines actually observed and unchanged; omit model-written descriptions. Return explicitly partial results when only some citations verify, and an error when none verify.
• Load MLX in disposable workers, serialize model loading across clients, and clean up after completion, cancellation, disconnect, or timeout.
• Refresh both skills with current lifecycle, query, setup, and recovery guidance. Updates refresh installed skills without deleting the venv.
• Add staged registration, connection, and inference diagnostics plus explicit launch-configuration repair.
• Pin the preserved upstream dependency to the exact previously installed revision and constrain the MCP SDK to its supported 1.x API.
• Repair Linux CI and release packaging; validate tag/package version consistency.

Validation: 125 tests pass, including in an isolated installation of the wheel; real MCP discovery and GPU inference pass. Read-only tests on artfct exercised successful source discovery, unsupported-result rejection, and worker cleanup.

Known limitation: broad exploration can still be slow or incomplete. Source verification establishes that cited lines were observed; callers should inspect their relevance and meaning.

v0.0.1

What's Changed

• Improve onboarding: agent-driven setup, prebuilt wheels, tests, CI by @rubybear-lgtm in #1

New Contributors

• @rubybear-lgtm made their first contribution in #1

Full Changelog: https://github.com/rubybear-lgtm/fastcontext/commits/v0.0.1