ContextBridge is a resume safety gate for AI coding sessions. When work moves between sessions, accounts or agents, it checkpoints the real state of the repo, records the tests you ran, and tells the next session whether it's safe to edit code.
Record
Run your tests through ContextBridge. The command, exit code and time are stored - whether it passes or fails.
Save
Checkpoint the branch, HEAD and file hashes as plain files, with a diff review before anything is saved.
Switch
Hand the work to another session, account or agent. The state travels with the repo, not the chat.
Resume
The next session runs the gate first. It reports READY, REVIEW or BLOCKED based on what changed since the checkpoint.
Figures from the project's own benchmarks: labelled drift scenarios, resume-check size, and a 10,000-file check. It also reports 99.9% agreement with an independent oracle across 800 random git histories. Zero runtime dependencies beyond Python's standard library and git.
Evidence
Records test commands, exit codes and timestamps, including the failures.
Checkpoints
Stores branch, HEAD and file hashes as plain files, with a diff review before saving.
Decision graph
Links requirements, decisions, code and tests, so the reason survives with the change.
Drift detection
Flags commits since the checkpoint, changed files and failed runs.
Secret redaction
Scans before writing. Best-effort - it reduces risk, it does not replace review.
Local-first and portable
JSON and Markdown files in your repo. No account, no telemetry.
Advisory session claims
Warns when two agents claim overlapping paths.
ContextBridge is free, MIT-licensed and sponsored by Orvantia. If you want AI-assisted work built into your team's process, that's what we do.