You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
This commit was created on GitHub.com and signed with GitHub’s verified signature.
Added
A tier-agent-spawns hook (PreToolUse on Agent) applies the delegation stance to every
subagent spawn that names no agent definition and no model, which is what a planning
framework's or a plugin's "launch a subagent" produces: one tier below the session under tiered, untouched under session-model, a prompt under off. Frontmatter and explicit
model choices are left alone. Inside a repository that carries a framework runtime
(_bmad/scripts/ or _bmad/core/), a bare spawn keeps the session model: the framework's
lenses are judgment work its overrides cannot rename, and its own rule is same capability.
templates/bmad/custom/ carries override files in BMad's own format for bmad-build, bmad-build-auto and bmad-code-review that run their review layers as the reviewer and spec-reviewer agents and the implementation handoff on the builder's tier. harness bmad apply installs them into a repository's _bmad/custom/, skipping a template whose keys or
layer ids the installed skill no longer declares, and harness bmad check reports that drift.
The harness-session hook runs apply for the repository a session starts in and says so
once when it wrote, kept or skipped a file.
docs/bmad.md records how the two pieces keep the delegation stance in charge of a
framework's spawns, and why the implementation handoff is not routed to builder.