restore-team-communications

name	restore-team-communications
version	0.2.0
description	Repair Claude teammate routing after same-session compaction or resume when sc-compose still exists on disk and the saved leadSessionId still matches the current SESSION_ID, but ATM or teammate reachability is broken.

Restore Team Communications

Use this skill only when all of these are true:

• ATM_IDENTITY=team-lead
• SESSION_ID still matches leadSessionId in ~/.claude/teams/sc-compose/config.json
• the team directory still exists on disk
• teammate communication is broken or suspect after compaction or resume

Do not use this skill for fresh startup or after clear. If the current SESSION_ID does not match leadSessionId, use /team-lead and follow the full restore procedure instead.

Step 0 — Prove Whether Repair Is Needed

First, try normal teammate communication before changing anything:

atm send quality-mgr "ping: verify sc-compose communications path"

If the message is delivered and acknowledged, stop. No repair is needed.

Do not read inbox files, session files, or tmux panes directly on this path.

Step 1 — Back Up Current State

Back up before any destructive recovery:

atm teams backup sc-compose
BACKUP_PATH=$(ls -td ~/.claude/teams/.backups/sc-compose/*/ | head -1)
cp -r ~/.claude/tasks/sc-compose/ "$BACKUP_PATH/tasks-cc"
echo "CC task list backed up to $BACKUP_PATH/tasks-cc"

Step 2 — Remove Broken Team Registration

Clear both live routing state and the persisted team directory:

TeamDelete

rm -rf ~/.claude/teams/sc-compose

If TeamDelete reports no active team name, proceed. That means live routing was already absent.

Step 3 — Recreate Team And Restore ATM State

Recreate the team:

TeamCreate(team_name="sc-compose", description="sc-compose development team", agent_type="team-lead")

Then restore from the most recent backup:

atm teams restore sc-compose --from "$BACKUP_PATH"

If required members are missing after restore, add them before verification.

Step 4 — Verify Both Communication Layers

Repair is not complete until all checks pass:

1. atm send quality-mgr "team restore verification ping".
2. atm send comp "team restore verification ping" and verify the Codex-side nudge or mailbox poll path still works.

For Codex-directed ATM sends, the nudge must include a clear call to action, not just a passive unread-mail announcement. Preferred structured nudge payload:

<atm><action>read atm</action><action>ack <TASK-ID></action><action>execute assigned task</action><when idle="immediate" busy="after-current-task"/><console announce="concise" pause="false"/></atm>

Fallback plain-text wording:

read atm for task <TASK-ID> and complete it before stopping

If the task is queued behind active work, use a queued-task nudge instead of an interruptive one.

Step 5 — Resume Work Quietly

If the repair succeeded:

• do not broadcast internal restore diagnostics over ATM
• send only the minimum teammate message needed to resume work
• return to normal project coordination

If the repair failed, stop and report the exact failed verification step.