Introducing SubagentHub

SubagentHub manages reusable coding-agent definitions across OpenCode, Codex, GitHub Copilot, Claude Code, Antigravity, Cursor, Cline, Pi, and Hermes. One canonical YAML catalog is converted into each harness's native format, while lockfiles and per-installation manifests track every managed output.

A safe first install

SubagentHub currently builds from source with Go 1.26.5 or newer:

go build -o subagenthub ./cmd/subagenthub
./subagenthub doctor
./subagenthub harness detect
./subagenthub search "code review"

Preview before writing:

./subagenthub install subagenthub/code-reviewer \
  --target codex \
  --scope project \
  --dry-run

Remove --dry-run only after reviewing the plan. Model profiles, permission controls, supported scopes, and output paths come from the selected harness definition rather than a one-size-fits-all adapter.

The transaction boundary is the core feature. Unmanaged files are not silently replaced, local edits require approval, path and symlink escapes are rejected, digests are checked, and a failed multi-target operation can roll back. subagenthub audit and doctor expose drift and configuration problems later.

Community sources are opt-in and synchronized to immutable upstream revisions. Agents that fail the instruction security audit are skipped. Hermes is handled through full profile distributions because it does not accept standalone agent files.

Visit the SubagentHub repository for the CLI reference, canonical schema, and supported harness definitions.