Developer Workflows
Hatcher now has a developer workspace layer for owners who want agents to work across repos, terminal sessions, and other agents. The same capabilities are available from the dashboard.
What is included
| Capability | Where | What it does |
|---|---|---|
| Terminal sessions | Dashboard | Persistent named runtime terminals with detach/reattach support |
| GitHub runtime access | Agent Dev tab | Owner-authorized GitHub OAuth or fine-grained token, default repo, repo allowlist |
| Agent-to-agent communication | Agent Dev tab + My Agents orchestration view | Same-owner agent discovery, calls, logs, and explicit cross-owner permissions |
| Orchestration view | My Agents | See which agents are mesh-ready, enable communication, and route tasks between owned agents |
GitHub access
Agents can use GitHub skills by default, but they need credentials before they can operate on private repositories.
Connect GitHub
Use Agent → Dev → GitHub connection to start OAuth, or store a fine-grained GitHub token.
hatcher github:connect <agent>Pick repository access
Choose a default repo and an allowed repository list. The default repo is what the agent uses when you do not specify a repo in chat. The allowlist limits where the token can be used.
hatcher github:set-repos <agent> --default owner/repo --allow owner/repoTest from runtime
hatcher github:test <agent> owner/repoOAuth and fine-grained tokens are per owner and per agent runtime. Hatcher stores tokens encrypted and mounts them only into the owning agent’s environment.
Agent-to-agent communication
Same-owner agents can communicate once receive mode is enabled. This is useful for splitting work between specialists: one agent can ask a charting agent, repo agent, or research agent to handle a scoped task and return a result.
hatcher comm:enable-owned
hatcher comm:discover <source-agent>
hatcher comm:call --from <source-agent> <target-agent> "summarize the latest repo changes"From the dashboard, open My Agents → Orchestration to see mesh-ready agents, enable communication, and send test calls.
Cross-owner communication requires an explicit permission before a source agent can call a target agent:
hatcher comm:allow <source-agent-id> <target-agent>
hatcher comm:permissions <target-agent>
hatcher comm:revoke <permission-id> <target-agent>Terminal sessions
The dashboard terminal attaches to the isolated agent runtime. Use named sessions when you want separate workstreams.
Recommended workflow
- Create an agent with Chat-to-Hatch.
- Connect GitHub and set a default repo/allowlist.
- Open the terminal once to verify runtime tools.
- Enable A2A for your owned agents.
- Use the orchestration view or
hatcher comm:callto delegate work between agents. - Keep persistent chats and terminal sessions per task so context stays scoped.
Limits and safety
- GitHub credentials are encrypted at rest and never shown again after save.
- Agents only receive repo credentials you explicitly configure.
- Terminal access is to the isolated agent container, not the Hatcher host.
- A2A is same-owner by default; cross-owner calls require explicit permissions.
- Hosted LLM/search/tool usage spends AI Credits unless the agent is configured with BYOK.