Problem
Today, Agent Zero is triggered through fairly rigid entry points: CLI flags (--feedback / --proactive), a labeled GitHub issue (gated by issues.requireLabel), or PR review comments. There's no lightweight, conversational way to just chat with the agent, describe what you want, and iterate on it in real time.
Proposed feature
Add a chat-style interface — similar to Blacksmith's Codesmith (chat via Slack, @codesmith-bot mentions on GitHub PRs, and a dedicated web UI) — that lets a user:
-
Dashboard chat panel — a persistent chat/session view in apps/dashboard where the user types a request, sees the agent's plan, actions, and evidence as it works, and can approve or redirect it, backed by the existing oRPC control plane (tasks.create, approvals.decide).
-
GitHub mentions — mention the bot directly in an issue or PR comment to start or continue a conversation bound to that task, without needing the issue to carry issues.requireLabel beforehand.
-
Persistent conversational sessions — the ability to keep chatting to refine scope or request follow-up changes without opening a brand-new issue each time.
Notes
This should stay consistent with Agent Zero's existing safety model — confidence gates, allowed change-risk class, repository permission, and verification/evidence — the chat is a new front door into the same policy-gated pipeline, not a bypass of it.
Problem
Today, Agent Zero is triggered through fairly rigid entry points: CLI flags (--feedback / --proactive), a labeled GitHub issue (gated by issues.requireLabel), or PR review comments. There's no lightweight, conversational way to just chat with the agent, describe what you want, and iterate on it in real time.
Proposed feature
Add a chat-style interface — similar to Blacksmith's Codesmith (chat via Slack, @codesmith-bot mentions on GitHub PRs, and a dedicated web UI) — that lets a user:
Dashboard chat panel — a persistent chat/session view in apps/dashboard where the user types a request, sees the agent's plan, actions, and evidence as it works, and can approve or redirect it, backed by the existing oRPC control plane (tasks.create, approvals.decide).
GitHub mentions — mention the bot directly in an issue or PR comment to start or continue a conversation bound to that task, without needing the issue to carry issues.requireLabel beforehand.
Persistent conversational sessions — the ability to keep chatting to refine scope or request follow-up changes without opening a brand-new issue each time.
Notes
This should stay consistent with Agent Zero's existing safety model — confidence gates, allowed change-risk class, repository permission, and verification/evidence — the chat is a new front door into the same policy-gated pipeline, not a bypass of it.