01 / Route
Pull eligible Linear work
Project routing rules pick the right workspace, issue, and execution context before an agent starts touching code.
Agentic Development Hub & Daemon
Talk is cheap, show me your agent system.
Turn Linear issues into an agent-driven engineering workflow: plan, implement, review, test, and loop with operators still in control.
run console
Current issue
Linear
ENG-123
Plan
strategy
Implement
worktree
Review
PASS
Daemon trace
01 route project: core-platform
02 attach skill: piv-implement
03 create draft PR: codex/eng-123
04 RESULT: PASS
Review contract
RESULT: PASS
SUMMARY: ready for merge
BUGS_JSON: []
How this works
01 / Route
Project routing rules pick the right workspace, issue, and execution context before an agent starts touching code.
02 / Plan
The planning stage builds a scoped implementation strategy from issue context, project config, and selected skills.
03 / Implement
The implementation agent works in the project workspace, updates code, and keeps pull request context synchronized.
04 / Verify
Structured review output feeds failures back into implementation until the run is done or clearly blocked.
Features
Route issues across many repositories without hard-coding behavior to a single workspace.
Comments, labels, status transitions, and issue state stay aligned with each workflow boundary.
Agents preserve pull request context and can update draft PRs as review feedback arrives.
Review parsing stays machine-readable with RESULT, SUMMARY, and BUGS_JSON outputs.
Run one issue, poll for new work, or let server-owned cron automation sweep eligible queues.
Humans can inspect stage, outcome, and risk while devos.ing handles the repetitive loop.
FAQ
devos.ing is ADHD: Agentic Development Hub & Daemon. It turns Linear issues into staged agent workflows for planning, implementation, review, and testing.
No. It removes repetitive coordination and execution loops while keeping operators in control of routing, review, and outcomes.
The current workflow centers on Linear, GitHub pull requests, CLI execution, server cron jobs, and agent adapters for Codex and Claude Code.
Yes. It can run one scoped issue, poll locally, or use scheduled server-owned automation sweeps across configured projects.
Review output follows a stable contract: RESULT, SUMMARY, and BUGS_JSON. Failures are fed back into implementation.
Run state is owned by the workflow layer inside the configured workspace, with safe recovery paths for partial runs.