Part of #656 (v0.10.0). Pipeline stage 4 of 4: hot-load / smooth use. This is the "make sure TMB can use it smoothly" key.
Goal
Once a resource is installed (#659), TMB uses it in the same session — ideally with no restart. If a restart is genuinely unavoidable, bro tells the user to run claude --resume and explains that TMB state is safe (trajectory-db checkpoints).
What to design
- In-session activation, by type:
- skill -> load/invoke it directly, or route the proper agent (e.g. swe, or a fresh consultant) to load it. Confirm a freshly-installed skill is discoverable without restart; if Claude Code only scans skills at session start, document that and use the resume fallback.
- toolkit / MCP server -> determine whether a newly-registered MCP server connects mid-session or requires a reconnect. If reconnect is needed, that is the restart fallback case.
- plugin -> enabling a plugin mid-session: test whether its hooks/MCP/commands load live or need a restart.
- Restart fallback (graceful): when live load is impossible, bro emits a clear instruction: "Installed . Restart needed to activate — run
claude --resume to continue exactly where we are (your tasks/decisions are checkpointed in the trajectory DB)." Verify --resume actually rehydrates TMB context (identity hook, in-flight issue, world model).
- Routing: after activation, bro picks who uses the resource (itself for planning context, or the executing agent) — the dashboard example should load the frontend skill into the agent doing the UI work.
Investigation needed (Claude Code behavior)
Document which resource types Claude Code can hot-load vs which require a restart. This determines how often the --resume fallback fires. (Use claude-code-guide / docs.)
Acceptance
For at least one resource type, a freshly-installed resource is used in the same session with no restart. For types that require a restart, bro gives the correct --resume instruction and resume is confirmed to preserve TMB state.
Part of #656 (v0.10.0). Pipeline stage 4 of 4: hot-load / smooth use. This is the "make sure TMB can use it smoothly" key.
Goal
Once a resource is installed (#659), TMB uses it in the same session — ideally with no restart. If a restart is genuinely unavoidable, bro tells the user to run
claude --resumeand explains that TMB state is safe (trajectory-db checkpoints).What to design
claude --resumeto continue exactly where we are (your tasks/decisions are checkpointed in the trajectory DB)." Verify--resumeactually rehydrates TMB context (identity hook, in-flight issue, world model).Investigation needed (Claude Code behavior)
Document which resource types Claude Code can hot-load vs which require a restart. This determines how often the
--resumefallback fires. (Use claude-code-guide / docs.)Acceptance
For at least one resource type, a freshly-installed resource is used in the same session with no restart. For types that require a restart, bro gives the correct
--resumeinstruction and resume is confirmed to preserve TMB state.