[FEATURE] User Interrupt Hook

Would be great if we could have hook on context limit (when auto-compact turned off)!

This issue has been inactive for 30 days. If the issue is still occurring, please comment to let us know. Otherwise, this issue will be automatically closed in 30 days for housekeeping purposes.

Would be nice to see this fixed for edge-triggered status / notification purposes

I need that too!

Consider me another vote for this.

I'm working on a utility that watches Claude activity and is effectively 'stuck' if it can't detect manual interuption.

I could use the Interrupted hook too 🙏

Very keen for this feature it is leaving my Claude code wrapper in very strange states.

I need this one too ;)

Pretty please 🙏 Thanks!

please

Would love to see this one too

Also shopping for this.

+1

Do not interrupt this issue !
I need it to ! ✨

Confirming this affects session status monitoring tools as well.

Use case: Agent Console Dashboard (ACD) — a TUI that tracks Claude Code session status by listening to hook events. When a session is active, ACD sets status to working. It relies on the Stop hook to transition the status back to attention (idle). Because Stop does not fire on user interrupt, any session where the user presses ESC or Ctrl+C is permanently stuck showing working until the next hook event fires.

Why workarounds fail for this case:

• Timeout: a session that is interrupted mid-task looks identical to one that is still running a long tool call
• PreToolUse heartbeat: only fires during tool execution, not during the initial inference phase when interrupts most commonly happen
• Process monitoring: fragile across macOS/Linux, breaks with remote sessions

Environment:

| Component | Version |
|-----------|---------|
| Claude Code | latest |
| Platform | macOS 15.x (Darwin) |

The Stop hook gap is documented in the official hook reference ("Does not run if the stoppage occurred due to a user interrupt") but there is no alternative hook that covers this case. A UserInterrupt or Stop:interrupted hook event would close this gap cleanly.

There's no native UserInterrupt hook yet, but you can approximate interrupt detection with existing hooks:

INPUT=$(cat)
REASON=$(echo "$INPUT" | jq -r '.stop_reason // "unknown"' 2>/dev/null)
HOOK_EVENT=$(echo "$INPUT" | jq -r '.hook_event_name // ""' 2>/dev/null)
MARKER="$HOME/.claude/.session-alive"
rm -f "$MARKER" 2>/dev/null
echo "$(date -Iseconds) clean-exit reason=$REASON" >> "$HOME/.claude/session-log.txt"
exit 0

MARKER="$HOME/.claude/.session-alive"
if [ -f "$MARKER" ]; then
    LAST=$(cat "$MARKER")
    echo "Previous session was interrupted (started: $LAST)" >&2
    echo "$(date -Iseconds) interrupted previous=$LAST" >> "$HOME/.claude/session-log.txt"
fi
echo "$(date -Iseconds)" > "$MARKER"
exit 0

date +%s > "$HOME/.claude/.heartbeat"
exit 0

External watcher (run in a separate terminal):

while true; do
    if [ -f "$HOME/.claude/.heartbeat" ]; then
        LAST=$(cat "$HOME/.claude/.heartbeat")
        NOW=$(date +%s)
        DIFF=$((NOW - LAST))
        if [ "$DIFF" -gt 30 ]; then
            echo "Claude interrupted or idle for ${DIFF}s"
        fi
    fi
    sleep 5
done

{
  "hooks": {
    "Notification": [{"matcher": "start", "hooks": [{"type": "command", "command": "bash ~/.claude/hooks/session-start-check.sh"}]}],
    "PostToolUse": [{"hooks": [{"type": "command", "command": "bash ~/.claude/hooks/heartbeat.sh"}]}],
    "Stop": [{"hooks": [{"type": "command", "command": "bash ~/.claude/hooks/interrupt-detector.sh"}]}]
  }
}

Not a perfect substitute for a native UserInterrupt event, but these patterns cover the main use cases: recovery after interrupt, logging interrupt vs clean exit, and real-time stall detection.

Very curious why this hasn't been implemented yet. It seems trivial to implement, and its absence plays havoc with tracking internal Claude state.

Throwing in another request for this. Seems like a pretty major hole in Claude Code hooks coverage, and an annoyance for multi-agent workflows

+1 — this gap affects out-of-process integrations badly.

I maintain a local sidecar service that mirrors Claude Code's session state
to a hardware status indicator (LED). Without a hook for user interrupts,
the indicator gets stuck in "processing" forever after the user presses Esc
— there's no event to clear it until the next UserPromptSubmit, which may
never come if the user walks away.

Inferring interrupts by tracking unpaired PreToolUse / PostToolUse
works for the "interrupt then continue" case but completely misses
"interrupt then idle/close window", and forces every integrator to
reimplement the same pairing logic.

A dedicated UserInterrupt (or Interrupted) hook event with session_id,
cwd, and the cancelled tool_use_id(s) would solve this cleanly without
overloading Stop's existing semantics. Would love to see this prioritized.

I need this feature added asap.
As Claude Code doesn't support any other file encoding other than UTF-8, I'm using hooks on Read/Edit/Write to convert my files to UTF-8, then parse the edited file back to my encoding. However, if I don't accept a change, I can't detect that the file needs to be converted back.

+1

+1

+1

+1

+1