Bring project memory to a coding agent
Give a coding agent useful project decisions through a brief or MCP connection without assuming it has your repository, secrets, or chat history.
Keep a project moving 2 min read
Resume a coding task in a new AI chat
Before leaving the old chat, record what changed, which checks actually ran, and what remains unverified. In the new chat, ask the agent to inspect the current working tree before using that checkpoint.
Preserve evidence, not a success story
“The feature is nearly done” does not help the next agent decide what to do. “The route compiles; the database test has not run” does. Include the exact check and its outcome when you have it. Do not turn an intended test into a passed test.
Note unrelated edits already present in the repository. A new agent should know not to overwrite them, stage them into a commit, or infer that they belong to the current task.
Inspect again after the handoff
A checkpoint is a record of a moment. Another contributor may have changed the branch since it was written. Ask the receiving agent to compare the note with the actual diff and recent commits.
Share only the logs needed to diagnose the remaining problem. Remove tokens, private payloads, and customer data before putting terminal output in a chat. A failing command and sanitized error are often enough to begin.
TAKE THIS WITH YOU
Coding checkpoint template
An illustrative template. Replace the details with material you are allowed to share.
Task: […] Branch / commit: […] Files changed: […] Behavior implemented: […] Checks run and actual results: […] Checks not run: […] Known failure: […] Unrelated edits to preserve: […] Next smallest action: […] Inspect the current checkout first. Do not assume this checkpoint describes the latest files.
USING THIS WITH DRAY
Keep the context. Check the source.
Stored project context can explain earlier decisions to a connected agent. Repository access and local transcript capture remain separate capabilities.
Dray is in beta. Available sources and connections depend on your setup. Read the data notes before granting access.
Sources & next reads
Product details are grounded in the guides below. Example projects and prompts are illustrative, not customer results.