Roadmap

Full roadmap at ROADMAP.md. The shape:

Shipped

• Framework substrate: preflight, lock, spend, claude_invoke, gh, slack, event-log, commit-trailer, handoff-table.
• launchd plist template + render.sh + deploy.sh.
• doctor.sh: fleet-wide preflight under HERMES_DOCTOR=1.
• hermes-claude: two-account swap helper.
• Issue claim state machine (agent:in-flight → agent:pr-open → agent:done) with race resolution + stale sweep.
• Slack severity routing (info / warn / alert).
• install.sh + INSTALL.md for fresh-machine bootstrap.
• Setup walkthroughs: Slack, AWS, Claude Code, skills, Linux stance, your-first-agent tutorial.
• Operator CLI: label-state + pre-push hook.
• CI (pytest + ruff + mypy + shellcheck + scrub-check) on every PR.
• Release automation (tag → GitHub release with auto-extracted changelog).
• Project hygiene: COC, security, support, issue templates, PR template, dependabot.
• pyproject.toml (ruff + mypy), pre-commit config.
• Homebrew formula pinned to the latest public release tarball.
• This Astro Starlight docs site.

In flight (next release)

• Bot token integration (xoxb-…). Unlocks slack_set_channel_topic(), threaded chat.postMessage for daily-thread routing of info-tier messages, reactions API.
• Drake-style proactive title-token dedup. Runner-level guard before invoking the planner.
• claim_pr / release_pr. Extend the state machine to PR-level work.
• render-systemd.sh. First-class Linux scheduling.
• Spend dashboards. Render a weekly recap from per-agent spend files.
• alfred new-codename scaffold. Single command to add a fresh codename agent.

Considered, not yet committed

• MCP server bundling (expose primitives as MCP tools).
• First-class GitHub App (vs operator’s PAT).
• Pluggable spend backends.
• Plugin system for skills.
• Web dashboard (rejected once; listed for visibility).

Out of scope (deliberately)

• Multi-tenant.
• Web UI.
• Long-running orchestration loop.
• Hosted model gateway.
• Browser automation built in.
• Vector DB for memory.
• Anything Anthropic ships natively (Agent Teams, Memory Tool, MCP server registry).
• Hosted SaaS.
• PyPI publishing.

Influence

• Strong: a working PR for a feature already on the in-flight list.
• Medium: a well-scoped feature request issue with a real use case + proposal.
• Low: “would be cool if” comments.
• None: scope-broadening requests (multi-tenant, hosted, web UI).