dsh-autopilot

Running a coding agent once is easy; operating one continuously is harder. Someone still has to decide which ticket is ready, track token spend against a budget, cap concurrency, recover interrupted work, publish the exact verified commit, and tell the team what happened. dsh-autopilot adds that operational layer to DeepSeek Harness, turning approved tracker tickets into bounded agent runs and review-ready pull requests, while humans keep review, CI decisions, and merge authority.

Core capabilities: only tickets carrying the configured ready marker and current human authorization are admitted; work is queued under schedule and concurrency limits with token-budget reservation reconciled against provider-reported usage; execution happens in a DSH-managed Git worktree with pause, resume, and cancel without losing durable run history; interrupted publication and notifications are recovered after restart; the exact Git state is revalidated before creating a pull request; a notification outage cannot undo a completed PR. Provider integration goes through official MCP servers with DSH Credentials reused, avoiding third-party clones.

Best for teams or individual developers already running coding agents on DSH and managing tickets in Jira or GitHub Issues, who want to automate the repetitive operational work between approval and delivery. It does not replace DSH or vendor integrations; it simply wires the scheduling, lifecycle, and publication behavior that joins them together.