wf-quit

wf-quit — Clean shutdown of the active workflow

This skill is invoked by /waterfall:quit. It stops the active team of the current session, cleans up session markers, and preserves the need's state for resumption via /waterfall:resume.

Step 1 — Detect the active session

Read the current session_id from the environment or the Claude Code session context.

Check whether the session marker exists:

~/.claude/wf-session-active.<session_id>

If the file does not exist: display the following message and stop (exit 0):

No active wf workflow in this session.

(EX-011 — no error, neutral behavior)

If the file exists: read its content to obtain the need_name:

need_name=$(cat ~/.claude/wf-session-active.<session_id>)

Step 2 — Load the team config

Build the config path:

~/.claude/teams/wf-<need_name>/config.json

Read the JSON to enumerate active teammates (field teammates or equivalent depending on harness format).

If the file is missing or corrupted: continue with an empty teammates list (R4 fallback — markers will be deleted anyway, teammates will time out on the harness side).

Step 3 — Send shutdown_request to each teammate

For each active teammate identified in step 2, send via SendMessage (plain text):

to: <teammate_name>
message: |
  type: shutdown_request
  request_id: wf-quit-<session_id>

Send messages in parallel if possible. Do not wait for a response (shutdown_response) — the shutdown is unilateral.

Step 3.bis — Close the active segment (ADR-008)

Before deleting markers, close the current segment in .wf-state.json:

state_file="wf/needs/$need_name/.wf-state.json"
source "${CLAUDE_PLUGIN_ROOT}/scripts/lib/wf-segments.sh"
_seg_close "$state_file" || true

This operation is atomic (jq + tmp + mv). If the file does not exist or jq fails, the error is silently ignored (robustness R4).

Step 3.ter — CronDelete watchdog if close_requested (EX-WCC-004)

If the watchdog wrote a close flag in status.json, wf-quit handles the CronDelete as a safety net (INV-WCC-002: wf-quit is on the HO side, therefore allowed to call CronDelete).

if ~/.claude/wf-watchdog-status.json exists:
  cron_id = jq -r '.cron_job_id // ""' ~/.claude/wf-watchdog-status.json
  close_req = jq -r '.close_requested // false' ~/.claude/wf-watchdog-status.json
  if close_req == "true" AND cron_id non-empty:
    CronDelete(cron_id)
    log cron_deleted_via_quit in or.log:
      bash ${CLAUDE_PLUGIN_ROOT}/scripts/wf-orchestrate.sh <name> --log \
        --msg '{ "ts": "<now>", "tag": "[WATCHDOG]", "event": "cron_deleted_via_quit", "cron_job_id": "<cron_id>" }'
else: silent skip

INV-WCC-003: do not delete ~/.claude/wf-watchdog-status.json — the file remains with status=OFF for statusline consistency. PM cleans the close_requested / cron_job_id fields via del() if it runs after.

Step 4 — Need-scoped marker cleanup (ADR-004)

Read the session_id from the need's .wf-state.json (not from the current env — avoids deleting markers from another window if /waterfall:quit is run out of context):

sid_besoin=$(jq -r .session_id "wf/needs/$need_name/.wf-state.json")
rm -f "$HOME/.claude/wf-session-active.$sid_besoin"

These deletions are:

• Idempotent: rm -f — no error if files already absent (EX-011)
• Cross-window safe: based on the sid stored in the state, not on $CLAUDE_SESSION_ID of the current window (ADR-004, INV-005)
• Non-destructive on .wf-state.json (EX-012 / INV-004)

Step 5 — Display confirmation

Display a confirmation message, for example:

Workflow wf-<need_name> stopped.
Teammates received shutdown_request.
Session markers deleted.

To resume: /waterfall:resume <need_name>

Rules

• ALLOWED only: close the active segment via _seg_close (step 3.bis) — the only mutation permitted on .wf-state.json
• FORBIDDEN: any other modification of .wf-state.json (phase, step, status, history — EX-012 / INV-004)
• FORBIDDEN: git commit, archiving, modification of SDD artifacts
• Idempotent: if called twice, no error (rm -f in the helper)
• Neutral if no session: informational message, exit 0 (EX-011)