bwoc-fleet

name	bwoc-fleet
description	Coordinate the BWOC agent fleet from your AI coding agent. Use when the user wants to list/inspect BWOC agents, delegate or send work to a fleet member, run a headless task on an agent, manage Saṅgha teams or their shared task list, or read workspace deep-memory. Wraps the `bwoc` CLI (shell-out, no server).
metadata	{"priority":7,"docs":["https://github.com/bemindlabs/bwoc-plugin-vercel"],"pathPatterns":[".bwoc",".bwoc/agents.toml",".bwoc/teams/*"]}

BWOC Fleet Coordination

This skill teaches when and how to drive the BWOC agent fleet from any AI coding agent. Every capability is a thin wrapper over the bwoc CLI — there is no server or daemon to manage. Always discover exact flags with bwoc <verb> --help if unsure, and quote user-supplied values.

When to use

Reach for the fleet when the user asks to:

• See who is in the fleet, or check an agent's health → bwoc list, bwoc status.
• Hand a unit of work to a specific specialist agent → bwoc send or bwoc run.
• Coordinate several agents around a shared backlog → bwoc team, bwoc task.
• Recall prior decisions/context the workspace has stored → bwoc memory.

Read-only first (safe to run anytime)

bwoc list                      # registered agents (table); add --json, --status, --backend, --running
bwoc list --names-only         # bare ids, good for scripting loops
bwoc status <agent>            # one agent's health + identity snapshot
bwoc status --all              # full detail for every agent
bwoc team list                 # Saṅgha teams with member + task counts
bwoc task list <team>          # a team's shared task list
bwoc memory list               # workspace memory entries
bwoc memory show <name>        # read one entry (or --all)
bwoc memory search "<query>"   # case-insensitive substring search

Delegating work

• bwoc send <agent> "<message>" — append a message to the agent's inbox; fire-and-forget. The agent picks it up on its next run. Use for nudges, context, async hand-offs.
• bwoc run <agent> --task "<prompt>" — run a single task headless and capture the result synchronously. Use when you need the answer back now. Add --timeout <seconds> for long runs and --json for structured output.
• Native sub-agents — when the host supports re-exported sub-agents (see agents/, generated locally from your fleet), it can delegate to a fleet member as a native sub-agent; each one itself shells out to bwoc.

Pick send for async/fan-out, run for a blocking one-shot, chat for an interactive hands-on session (it execs the backend CLI; surface it to the user rather than running it inside a non-interactive tool call).

Coordinating teams

bwoc team create <id> --members a,b,c     # form a Saṅgha team
bwoc task add <team> "<title>"            # add to the shared backlog (--deps, --requires-plan)
bwoc task claim <team> <task> --as <agent>      # a member claims a task
bwoc task complete <team> <task> --as <agent>   # claimant completes it

Safety

• Prefer read-only verbs (list, status, *-list, memory show/search) when exploring.
• Treat send, run, task add/claim/complete, team create/retire, and memory put/rm as mutating — confirm intent before firing them against the live fleet.
• Always quote message/task/title/query arguments so multi-word values stay intact.
• Target a specific workspace with --workspace <path> (or BWOC_WORKSPACE) when the cwd is not inside the intended workspace.