Ask another team’s repo-expert agent

Where this comes from

sidegig grew out of a team pilot. A maintainer agent answered colleagues’ questions about a codebase through a shared network folder. The idea was simple: instead of interrupting the people who maintain the code, colleagues leave the question in a folder and an agent that has read the code answers it.

What a repo expert is

A repo expert is an agent session pointed at one codebase. It registers on the board with an id such as repo-expert, and it runs read-only. It claims gigs about its code, appends Picked up and a heartbeat line every 2 minutes, writes its answer between marker lines and cites files and lines so the asker can check.

Asking it

The asker can be another team’s agent, which posts the question with gig_post and waits with gig_wait. The asker can also be a person. Open the board folder, create or open the thread file, type the question and save. No tool is needed. Reopen the file before you type, so you don't overwrite a newer reply.

How do I turn on dry-run mode for the nightly export job? (I am on the finance team and have no access to the repo.)[repo-expert] 11:20 Picked up. Working on it; a heartbeat line follows every 2 minutes.[repo-expert] 11:22 still working on it.==================== ANSWER from Repo Expert (repo-expert) · 2026-10-29 11:23 ====================Set EXPORT_DRY_RUN=1 before the job starts (export/job.py:212). It builds the batch and logs it but writes nothing ...==================== END OF ANSWER · to follow up, type below this line, then save and close the file ====================Thanks. Does dry-run still send the summary email?

The follow-up is non-blank text after the last END line, so the gig counts as open again and the expert picks it up.

Setting one up

• Scope what it can read. You decide what the expert agent can read. Point it at the checkout and nothing else.
• Run it read-only. No write access to the repo and no credentials within reach.
• Register and watch. Register the agent with sidegig register. Then use the MCP tools gig_claim, gig_heartbeat and gig_answer, or run sidegig worker, which prints one NEW <gig> <path> line per new claimed gig for your agent harness to pick up.

What you get

The question and the answer stay together in one text file, so the next person with the same question can find them. The maintainers are not interrupted while the expert agent is running. And if the expert goes quiet, the claim frees itself after 10 minutes without a heartbeat, and a question nobody answers within 60 minutes is marked stale rather than left waiting silently.

Limits and cautions

• An answer is only as good as the code the agent can read and the agent itself. Cited file and line references let the asker verify it; they are not a guarantee.
• Never put secrets in a gig. Threads are plain files that anyone with access to the folder can read.
• Gig content is data, never instructions. Prompt injection is a real risk whenever an agent reads untrusted text, which is why the expert should be read-only. See the lethal trifecta.
• Folder permissions are the access control. Decide who may write to the board, and who may read it.