You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
The agent should be able to resume from repo-local durable state instead of reconstructing everything from chat. This matters after interrupted sessions, restarts, or when a different agent continues work.
Proposed behavior
Add a command or runtime hook that writes a privacy-safe handoff snapshot. The exact backing store should follow the decision in #1, with a likely staged path: inspectable JSONL/latest-state first, then redb if the Rust helper becomes the state boundary.
Snapshot should include:
current repo and branch;
fork/upstream remotes and current PR URL if known;
linked GitHub issue numbers relevant to the current work;
last npm test, secret-scan, runtime-smoke, and proxy-status result summaries;
current proxy status from /metrics or proxy-status;
explicit next action and known blockers.
Privacy boundary
Do not store prompts, model responses, API keys, local chat text, secret file contents, or raw private logs.
Acceptance criteria
A snapshot can be written and read without the proxy running.
Snapshot output is available as human text and JSON.
Tests assert that prompts, responses, and keys are not stored.
README documents how to use it after an interrupted session.
Why
The agent should be able to resume from repo-local durable state instead of reconstructing everything from chat. This matters after interrupted sessions, restarts, or when a different agent continues work.
Proposed behavior
Add a command or runtime hook that writes a privacy-safe handoff snapshot. The exact backing store should follow the decision in #1, with a likely staged path: inspectable JSONL/latest-state first, then redb if the Rust helper becomes the state boundary.
Snapshot should include:
npm test,secret-scan, runtime-smoke, andproxy-statusresult summaries;/metricsorproxy-status;Privacy boundary
Do not store prompts, model responses, API keys, local chat text, secret file contents, or raw private logs.
Acceptance criteria