Imported from joshka0/foxctl (configs/skills-pack/foxctl-room-operator/SKILL.md). Install upstream with npx skills add joshka0/foxctl --skill foxctl-room-operator. Copyright stays with the author.

foxctl-room-operator

Use this skill when you are already working inside an existing foxctl room and need to behave correctly as a participant, reviewer, or coordinator.

This skill is about room operating protocol, not room creation.

If the room is explicitly running the agile epic/milestone/story workflow, also use foxctl-room-agile.

Use me when

• a room already exists and you need to pick up or continue work
• you were assigned a room task
• you need to reply to a direct room request
• you are acting as coordinator or reviewer in a live room
• you need to know whether to send, ack, resolve, claim, touch, block, or complete
• you need to answer or verify a room interview question
• you need to manage a bounded scheduled follow-up for a participant
• you need to work inside an epic/milestone/story structure instead of only tasks and chat

Core rules

• foxctl is the headless room/runtime kernel.
• CLI, web/API, gui-agent, mux panes, and chat adapters are clients or presentation layers over that kernel.
• The room timeline is canonical. Pane scrollback is not.
• Participant transport and mux presentation are different things. A pane label or mux session is not proof of live delivery.
• Room-scoped clients should prefer GET /api/rooms/{room-id}/events?workspace_id=... over the global /api/events feed for room timeline updates.
• Room member delivery_binding is the canonical transport/routing record. Older top-level member transport fields are not emitted in room member API responses.
• Reply-required work is chain-aware: a later message only satisfies an earlier request when it is from the intended recipient and belongs to the same room message chain.
• Room status and loop surfaces expose a durable last_delivery_trace: use that trace to debug the last chosen binding, transport, fallback attempt, and outcome instead of guessing from pane behavior.
• use explicit, distinctive participant ids.
• avoid generic actor ids like coordinator, reviewer, or codex-coordinator when multiple rooms or agent runtimes may coexist.
• actor ids should include feature, branch, room, or workstream context, for example feat-room-loop-codex, room-ci-reviewer-a, or runtime-owner-a.
• Room context and room membership are different things. FOXCTL_ROOM_ID, a tmux/zellij session name, or a startup prompt only mean the pane knows about the room. They do not prove the participant is joined and routable.
• Room durability and live execution are different things. A room can continue to hold epic/task/history state even when there are no panes or live participant runtimes.
• When invoking room send, prefer explicit --to <participant> for anything meant for one person (so relay + inbox stay aligned), and --sender <you> whenever your pane context might not identify you (scripts, MCP, or outside tmux/zellij). Omit --to only for deliberate broadcasts to the rest of the room.
• Start with room status, then room inbox --actor <you>.
• Direct obligations beat casual browsing of the timeline.
• If a task is assigned to you, claim it before doing real work.
• When creating a new room task, treat lane selection as explicit contract:
  • plain room task add defaults to the newest open epic's quiet Chores milestone when one exists
  • use room task add --milestone-id <milestone-id> for milestone-scoped execution work
  • do not use title keywords to decide where a task belongs
• If work takes time, touch the task instead of posting vague status chat.
• If blocked, room task block --reason ... with a concrete reason.
• When done, room task complete --notes ... so the outcome is durable.
• Use room send --to @coordinator for escalation or routing decisions.
• Do not work a task that another participant already owns unless the coordinator reassigns or reclaims it.
• If the room is using the interview protocol, use room interview next --actor <you> before browsing the full transcript.
• If the room is using the agile layer, orient from room epic show, room milestone show, or room story show instead of reconstructing the tranche from raw chat.
• Read the participant state shown by room status as four separate dimensions: membership, transport availability, runtime availability, and presentation attachment.

Room entry flow

Run this first:

foxctl room status <room-id>
foxctl room inbox <room-id> --actor <you>
foxctl room task list <room-id>
foxctl room interview next <room-id> --actor <you>

If the room is using the agile layer, orient here too:

foxctl room epic show <room-id>
foxctl room milestone show <room-id>
foxctl room story show <room-id>

Practical rule:

• if an epic is still in discovery or intake_in_progress, do not start milestone work yet
• answer epic intake questions first
• milestones open only after the coordinator finalizes the epic brief
• after finalization, prefer foxctl room epic shape <room-id> <epic-id> so milestone proposals become durable room objects before execution starts
• when the coordinator opens a milestone from one of those proposals, prefer foxctl room milestone start <room-id> <epic-id> --proposal <proposal-id> over retyping the proposal by hand

If you are in an existing zellij pane and the environment is missing FOXCTL_ROOM_ID / FOXCTL_PARTICIPANT, you are not room-bound yet even if the pane lives in the same zellij session as other room participants. Bind the current pane explicitly before assuming relay or coordinator messages will land:

foxctl room join <room-id> --current --role <room-role>

If the participant already exists in the room but moved to a different tmux/zellij pane, update the transport binding instead of re-joining:

foxctl room rebind <room-id> <actor-id> --backend tmux --session <session> --pane-id <pane>

Practical rule:

• tmux or zellij session membership is not the same thing as room membership
• if room status does not show your participant, you are not joined yet even if the pane has room env vars or a room startup note
• each pane that should receive room traffic must either join explicitly or auto-register transport when launched by foxctl
• do not assume a session-wide broadcast will reach unmanaged panes
• task assign/claim/complete use the same transport-first relay path as room relay; pane delivery is a viewer effect, not the room’s source of truth
• if no agents are meant to execute right now, it is valid to leave the room in durable-room-only mode and avoid rebuilding panes until they are actually needed
• operator surfaces must converge on the same shared runtime truth; if CLI, API, and GUI disagree, treat that as a room-runtime bug, not normal behavior

Then decide:

• direct ask waiting on you: answer or acknowledge it first
• interview item waiting on you: answer or verify it first
• assigned task waiting on you: claim it
• no direct ask, no assigned task: ask the coordinator or pick unclaimed work only if the room policy allows it

Correct action by situation

I need to answer a direct request

Use:

foxctl room send <room-id> --to <sender> --sender <you> "Your reply"

If the request only needs confirmation:

foxctl room ack <room-id> <message-id>

I am starting assigned work

Use:

foxctl room task claim <room-id> --id <task-id>

I need to create new work

Use plain task creation for quiet chores:

foxctl room task add <room-id> --title "Follow up on reviewer debt"

Use explicit milestone linkage for active delivery work:

foxctl room task add <room-id> \
  --title "Implement current slice" \
  --milestone-id <milestone-id>

If you do not know the correct milestone, inspect room milestone show <room-id> first.

I am still working and want to keep heartbeat current

Use:

foxctl room task touch <room-id> --id <task-id>

I am blocked

Use:

foxctl room task block <room-id> --id <task-id> --reason "waiting on <specific thing>" --sender <you>
foxctl room send <room-id> --to @coordinator --sender <you> --reply-expected "Blocked on <specific thing>"

I finished the task

Use:

foxctl room task complete <room-id> --id <task-id> --notes "<completion note>"

I need a coordinator decision

Use:

foxctl room send <room-id> "Need coordinator input on <issue>" --to @coordinator --reply-expected

Prefer --sender <you> whenever you are not obviously in the correct room-bound pane.

I need a durable scheduled follow-up

Use:

foxctl room remind add <room-id> <participant> "Check MR !26 and report status" \
  --every 15m \
  --max-iterations 3 \
  --reply-expected

This writes the original direct request now, then lets the room loop send bounded reminder follow-ups until the recipient replies or the retry budget is exhausted.

Operator rule:

• if you set or expect reminders, confirm room loop is running for that room
• room relay alone is only pane fanout; it will not drive reminder follow-ups or stale-reply nudges
• room loop reads persisted policy; do not expect runtime flags to change reminder or stale-work behavior
• task_followup_interval=0 disables automatic task follow-up check-ins while leaving reminder follow-ups active
• acknowledging one emitted reminder instance does not stop the recurring schedule
• recurring reminders can also stop when linked task_id, story_id, or milestone_id work is completed

I need to understand the current milestone

Use:

foxctl room milestone show <room-id>
foxctl room story show <room-id>

Read the milestone acceptance criteria before starting a story. If the criteria are unclear, use the room interview or planning protocol instead of guessing.

I am the interview respondent

Use:

foxctl room interview next <room-id> --actor <you>
foxctl room interview answer <room-id> <question-id> "<answer>"

Do not answer by free-form room chat if the room is already using a durable interview session.

I am the interview verifier

Use:

foxctl room interview next <room-id> --actor <you>
foxctl room interview verify <room-id> <answer-id> accept "<why it matches>"
foxctl room interview verify <room-id> <answer-id> clarify "<what is still ambiguous>"
foxctl room interview verify <room-id> <answer-id> reject "<why it diverged>"

The verifier is usually the original submitter or the coordinator.

I am the coordinator

Use these as your primary controls:

foxctl room status <room-id>
foxctl room resolve <room-id> <message-id> --mode read
foxctl room task assign <room-id> --id <task-id> --to <participant>
foxctl room task reassign <room-id> --id <task-id> --to <participant> --reason "<reason>"
foxctl room task reclaim <room-id> --id <task-id> --reason "<reason>"
foxctl room coordinator set <room-id> <participant>

Coordinator responsibility:

• keep assignments explicit
• keep stale work moving
• close reminder noise when it has already been handled
• make the final call on routing and review closure
• distinguish transport failures from viewer-only problems before escalating relay bugs
• own coordinator-only mutation surfaces: full room patching, full member replacement, and role-changing member binding updates should not be delegated as self-service room actions

I am the reviewer

Default reviewer behavior:

• findings first
• then verdict: approved or blocked
• then scope and any non-blocking follow-ups

Do not leave review conclusions only in pane chat. Write them into the room or task notes.

Completion and review note templates

Implementation/completion notes should include:

• changed: what changed
• verified: tests/build/manual checks run
• remaining: known gaps or follow-ups

Example:

changed: wired loop PATCH and persisted runtime state
verified: make test-short; npm --prefix packages/gui-agent run build
remaining: local GUI auth still uses dev-local-user fallback

Review notes should include:

• result: approved or blocked
• findings: count and severity
• scope: files/components/behavior reviewed

Example:

result: approved
findings: 0 blocking, 1 non-blocking follow-up
scope: /loop API, coordinator gating, reminder floor behavior

Anti-patterns

Do not:

• treat pane scrollback as the room source of truth
• post “working on it” repeatedly instead of task touch
• silently start assigned work without claim
• resolve coordinator-only items if you are not acting with coordinator authority
• close a review gate without an explicit approved or blocked outcome
• treat a rejected task transition as a bug before you check the task lifecycle: assign, reassign, claim, block, unblock, complete, reclaim, and abandon now enforce a stricter state machine

Verification

• Prefer bash tests/regression/run.sh as the canonical regression entrypoint when you need to re-check the hardened room-runtime invariants end to end.

Related

• configs/skills-pack/foxctl-room/SKILL.md
• configs/skills-pack/foxctl-room-agent/SKILL.md
• configs/skills-pack/foxctl-room-view/SKILL.md
• docs/general/tmux-collaboration.md