cancel

Cancel Skill

Intelligent cancellation that detects and cancels the active OMC mode.

The cancel skill is the standard way to complete and exit any OMC mode. When the stop hook detects completion, it instructs the LLM to invoke this skill for state cleanup. Retry a failed operation in the same scope and report what remains; --force changes shutdown urgency, never scope, lock handling, or ownership checks. --all is the only cross-session authorization.

Cancellation Contract

The flags control independent scope and shutdown dimensions:

Resolve the current identity from OMC_SESSION_ID in CLI contexts or the hook-provided session_id. A missing or ambiguous identity for no flags or --force fails closed: do not aggregate state, choose a discovered session, or issue an unscoped cleanup. This is a fail-closed error for the default command and for --force. --all may enumerate known sessions, but every concrete id is still read, cancelled, and cleared independently.

What It Does

Automatically detects and cancels active modes:

• Autopilot: Stops the workflow while preserving progress for resume.
• Ralph: Stops the persistence loop.
• Legacy Ultrawork / UltraQA: Removes stale upgraded-installation state only; neither is a live workflow.
• Ultragoal: Clears its session runtime guard while preserving durable .omc/ultragoal/ plan and ledger artifacts.
• Swarm: Stops coordinated agents and releases claimed tasks only in its authorized shared cleanup path.
• Ultrapilot / Pipeline: Stops their worker or sequential execution.
• Team: Uses the active team surface for graceful shutdown, then clears session-scoped Team state; forced paths skip waits. Claude Code 2.1.178+ has no TeamDelete.
• Team + Ralph: Team is cancelled first, then linked Ralph is cleared in the same session; a Team failure retains linked Ralph.
• OMC Teams, Plan Consensus, and Self-Improve: Use their mode-specific protected cleanup and preserve artifacts designated for resume.

Usage

/oh-my-claudecode:cancel
/oh-my-claudecode:cancel --force
/oh-my-claudecode:cancel --all
/oh-my-claudecode:cancel --force --all

Or say: cancelomc, stopomc.

Critical: Deferred Tool Handling

The state tools (state_clear, state_read, state_write, state_list_active, state_get_status) may be deferred by Claude Code. Before calling any of them, load all of them with ToolSearch:

ToolSearch(query="select:mcp__plugin_oh-my-claudecode_t__state_clear,mcp__plugin_oh-my-claudecode_t__state_read,mcp__plugin_oh-my-claudecode_t__state_write,mcp__plugin_oh-my-claudecode_t__state_list_active,mcp__plugin_oh-my-claudecode_t__state_get_status")

If state_clear is unavailable or fails, retry the same scoped operation and report the failure. The fallback below is only an emergency escape from a stop-hook loop for a non-force, non-Team cancellation. It removes files only for an explicitly identified current session. Do NOT use this fallback for autopilot, team, or omc-teams, or for --force/--all; never use it to bypass a lock, ownership check, corruption, or another state-tool failure. Linked modes need one protected operation per mode.

Replace MODE with the specific mode (for example ralplan, ralph, ultrawork, or ultragoal).

# Emergency fallback: direct file removal for one identified session only.
SESSION_ID="${OMC_SESSION_ID:-${CLAUDE_SESSION_ID:-${CLAUDECODE_SESSION_ID:-}}}"
REPO_ROOT="$(git rev-parse --show-toplevel 2>/dev/null || { d="$PWD"; while [ "$d" != "/" ] && [ ! -d "$d/.omc" ]; do d="$(dirname "$d")"; done; echo "$d"; })"
sha256portable() { printf '%s' "$1" | (sha256sum 2>/dev/null || shasum -a 256) | cut -c1-16; }

if [ -n "${OMC_STATE_DIR:-}" ]; then
  SOURCE="$(git remote get-url origin 2>/dev/null || echo "$REPO_ROOT")"
  HASH="$(sha256portable "$SOURCE")"
  DIR_NAME="$(basename "$REPO_ROOT" | sed 's/[^a-zA-Z0-9_-]/_/g')"
  OMC_STATE="$OMC_STATE_DIR/${DIR_NAME}-${HASH}/state"
elif [ "$REPO_ROOT" != "/" ] && [ -d "$REPO_ROOT/.omc" ]; then
  OMC_STATE="$REPO_ROOT/.omc/state"
else
  echo "ERROR: Could not locate .omc state directory" >&2; exit 1
fi
[ -d "$OMC_STATE" ] || { echo "ERROR: State dir not found at $OMC_STATE" >&2; exit 1; }
MODE="ralplan" # replace with the target mode
[ -n "$SESSION_ID" ] || { echo "ERROR: current session identity is required; refusing unscoped fallback" >&2; exit 1; }

if [ -d "$OMC_STATE/sessions/$SESSION_ID" ]; then
  rm -f "$OMC_STATE/sessions/$SESSION_ID/${MODE}-state.json" \
    "$OMC_STATE/sessions/$SESSION_ID/${MODE}-stop-breaker.json" \
    "$OMC_STATE/sessions/$SESSION_ID/skill-active-state.json"
  NOW_ISO="$(date -u +"%Y-%m-%dT%H:%M:%SZ")"
  EXPIRES_ISO="$(date -u -d "+30 seconds" +"%Y-%m-%dT%H:%M:%SZ" 2>/dev/null || python3 - <<'PY'
from datetime import datetime, timedelta, timezone
print((datetime.now(timezone.utc) + timedelta(seconds=30)).strftime('%Y-%m-%dT%H:%M:%SZ'))
PY
)"
  printf '{"active":true,"requested_at":"%s","expires_at":"%s","mode":"%s","source":"bash_fallback"}' \
    "$NOW_ISO" "$EXPIRES_ISO" "$MODE" > "$OMC_STATE/sessions/$SESSION_ID/cancel-signal-state.json"
fi

Cancellation Flow

This is the authoritative procedure for every invocation.

1. Parse Arguments

Parse exact flags independently; unknown or near-match arguments do nothing:

FORCE_MODE=false
ALL_MODE=false
for arg in "$@"; do
  case "$arg" in
    --force) FORCE_MODE=true ;;
    --all) ALL_MODE=true ;;
  esac
done
SCOPE="current"
if [[ "$ALL_MODE" == true ]]; then SCOPE="all"; fi
GRACEFUL_MODE=true
if [[ "$FORCE_MODE" == true ]]; then GRACEFUL_MODE=false; fi

The parsed values select exactly one row in the contract above.

2. Detect Active Modes

1. For no flags or --force, require one unambiguous current session id and call state_list_active(session_id="<current_session_id>"). Its scoped result is authoritative; never use an aggregate result to infer ownership.
2. For --all, call state_list_active(all=true) once and retain every concrete session id. Discovery enumerates targets; it is not a Team read.
3. For each selected id, call state_get_status(mode="<mode>", session_id="<session_id>") before mode cleanup. Treat Ultrawork as retired cleanup input.
4. Pass the selected id to every read, write, and clear. Protected state tools enforce the shared mutation lock and ownership boundary; no flag bypasses either one.

Use this dependency order when several modes appear in one scoped status:

1. Autopilot (including its linked Ralph and retired-state cleanup)
2. Standalone Ralph
3. Retired Ultrawork and UltraQA
4. Ultragoal runtime guard
5. Shared Swarm state (global pass only)
6. Ultrapilot and Pipeline
7. Native Team
8. Legacy OMC Teams
9. Plan Consensus and Self-Improve

Ralph linked to Team is excluded from the standalone step and is handled only after Team succeeds; this prevents the persistence loop from outliving workers.

3. Cancel Each Selected Session

Process each selected session independently in this dependency order. A failed or unresolved scoped operation marks that session unsuccessful: retain affected state/runtime records, report the scope and reason, and do not claim linked cleanup. An --all run may continue other ids, but any such failure blocks the final global pass.

Autopilot (primary first)

1. Read the exact session state, including workflowRunId when present.
2. Pause only that run with state_write(mode="autopilot", session_id="<session_id>", active=false, state={workflowRunId: "<exact run id>"}); do not replay the readback. The state tool revalidates workflow integrity under its mutation lock.
3. If the primary write fails, stop this session's dependent cleanup. Otherwise, clear only same-session named-workflow state_clear(mode="ralplan", session_id="<session_id>"), then linked Ralph, then retired UltraQA. Preserve the paused primary for resume when a dependent clear fails. Force mode keeps this primary-first ordering.

The primary pause preserves workflow, pipeline tracking, and task identity. A target_state_sha256 may be supplied only when it is the exact hash of the current serialized state. Never clear nested ralplan, linked state, cancel signals, or runtime artifacts after a failed primary write; report the exact dependent failure and retry it in the same session.

Team (native Claude Code)

1. Read state_read(mode="team", session_id="<session_id>") and obtain the captured team name and active worker labels. Never select a Team from an aggregate read.
2. On graceful paths, signal each named teammate through the active surface, wait up to 15 seconds per member, then reconcile for 5 more seconds. Record acknowledgements and timeouts. Forced paths skip both waits.
3. After graceful confirmation/timeouts (or immediately when forced), clear state_clear(mode="team", session_id="<session_id>"). If the protected clear fails for lock, ownership, corruption, or another reason, retain Team runtime records and linked Ralph; do not use an unscoped clear or fallback.
4. Only after Team clear succeeds, inspect linked Ralph and clear it with the same session id (state_read(mode="ralph", session_id="<session_id>") then `state_clear(mode="ralph", session_id="<session_i

Content truncated.