How do I install the Check On Task skill?

Run npx skillmds@latest add colindmurray/check-on-task in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Check On Task safe to use?

This skill has not completed SkillMD's automated safety review yet. Independent scanners report: SkillSpector: PASS, Skill Scanner: PASS. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Check On Task?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Check On Task free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Check On Task?

colindmurray (@colindmurray) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Check On Task by colindmurray — Agent Skill

Name: Check On Task
Author: colindmurray

File contents

Check on Task

Check the status of ANY background task launched by execute_long_running_task.

Usage

# Check a specific task
~/.openclaw/skills/long-running-task/bin/check_task --task-id <TASK_ID>

# Check all tasks for a session
~/.openclaw/skills/long-running-task/bin/check_task --session-id <SESSION_ID>

# Check all active tasks
~/.openclaw/skills/long-running-task/bin/check_task --all

# JSON output (for programmatic use — includes system memory info)
# Output shape: {"tasks": [{...}, ...], "system": {...}}  — task array is wrapped under "tasks"
~/.openclaw/skills/long-running-task/bin/check_task --all --json

# Custom stall threshold (default 600s = 10 min)
~/.openclaw/skills/long-running-task/bin/check_task --task-id <ID> --stall-threshold 300

Status Meanings

Status	Meaning
running	PID alive, output flowing normally
stalled	PID alive but no output for >stall-threshold seconds
completed	Process exited with code 0
failed	Process exited with non-zero code
dead	PID not found (auto-moved to failed)

Health Indicators

Health	Meaning
ok	Everything normal
stalled	Output file not written to recently
hang-detected	(Coding agents) Result event received but PID still alive — safe to kill
error	Process failed or PID dead

Monitor Watchdog

Each background task launched with --mode heartbeat gets a monitor watchdog (monitorPid):

Runs as a sibling process that survives if the task dies (OOM, SIGKILL)
Checks PID liveness every 120 seconds
If the task dies unexpectedly, the monitor:
- Moves the manifest to failed/ with monitorDetected: true
- Sends a notification immediately
- Fires a system event to wake the AI for retry decisions

Interpreting Monitor Fields

Field	Meaning
`monitorPid`	PID of the watchdog process
`monitorPid (alive)`	Monitor is actively watching the task
`monitorPid (dead)`	Monitor died — task has no fast-path death detection
`monitorDetected: true`	Task death was caught by monitor (fast-path, <2 min)
No `monitorDetected`	Task death was caught by cleanup_tasks (slow-path, up to 30 min)

If monitorDetected: true, the user has already been notified — focus on interpretation and retry decisions, not re-notification.

For Coding Agents

When the task type is coding-agent, you get deeper analysis:

Tool counts: how many Read/Edit/Bash tools the agent has used
Error count: error events in the NDJSON stream
Phase estimation: exploring (read-heavy), implementing (edit-heavy), testing (bash-heavy)
Git progress: commits since task start, unstaged changes
Hang detection: Claude stream-json bug where process stays alive after result event
System memory: current memory usage (useful for OOM risk assessment)

Caution on Killing

Only kill a process if it is clearly stuck (completely silent output for extended period, error loops, or hang detected with result event + PID alive). Coding agents often have long pauses between tool calls — this is normal.

---name: check-on-taskdescription: Check status of background tasks and coding agents — discover running processes, detect stalls, analyze progress, and generate clear reports.---# Check on TaskCheck the status of ANY background task launched by `execute_long_running_task`.## Usage```bash# Check a specific task~/.openclaw/skills/long-running-task/bin/check_task --task-id <TASK_ID># Check all tasks for a session~/.openclaw/skills/long-running-task/bin/check_task --session-id <SESSION_ID># Check all active tasks~/.openclaw/skills/long-running-task/bin/check_task --all# JSON output (for programmatic use — includes system memory info)# Output shape: {"tasks": [{...}, ...], "system": {...}} — task array is wrapped under "tasks"~/.openclaw/skills/long-running-task/bin/check_task --all --json# Custom stall threshold (default 600s = 10 min)~/.openclaw/skills/long-running-task/bin/check_task --task-id <ID> --stall-threshold 300```## Status Meanings| Status | Meaning ||--------|---------|| **running** | PID alive, output flowing normally || **stalled** | PID alive but no output for >stall-threshold seconds || **completed** | Process exited with code 0 || **failed** | Process exited with non-zero code || **dead** | PID not found (auto-moved to failed) |## Health Indicators| Health | Meaning ||--------|---------|| **ok** | Everything normal || **stalled** | Output file not written to recently || **hang-detected** | (Coding agents) Result event received but PID still alive — safe to kill || **error** | Process failed or PID dead |## Monitor WatchdogEach background task launched with `--mode heartbeat` gets a **monitor watchdog** (`monitorPid`):- Runs as a sibling process that survives if the task dies (OOM, SIGKILL)- Checks PID liveness every 120 seconds- If the task dies unexpectedly, the monitor: - Moves the manifest to `failed/` with `monitorDetected: true` - Sends a notification immediately - Fires a system event to wake the AI for retry decisions### Interpreting Monitor Fields| Field | Meaning ||-------|---------|| `monitorPid` | PID of the watchdog process || `monitorPid (alive)` | Monitor is actively watching the task || `monitorPid (dead)` | Monitor died — task has no fast-path death detection || `monitorDetected: true` | Task death was caught by monitor (fast-path, <2 min) || No `monitorDetected` | Task death was caught by cleanup_tasks (slow-path, up to 30 min) |If `monitorDetected: true`, the user has already been notified — focus on interpretation and retry decisions, not re-notification.## For Coding AgentsWhen the task type is `coding-agent`, you get deeper analysis:- **Tool counts**: how many Read/Edit/Bash tools the agent has used- **Error count**: error events in the NDJSON stream- **Phase estimation**: exploring (read-heavy), implementing (edit-heavy), testing (bash-heavy)- **Git progress**: commits since task start, unstaged changes- **Hang detection**: Claude stream-json bug where process stays alive after `result` event- **System memory**: current memory usage (useful for OOM risk assessment)## Caution on KillingOnly kill a process if it is clearly stuck (completely silent output for extended period, error loops, or hang detected with result event + PID alive). Coding agents often have long pauses between tool calls — this is normal.

Frequently asked questions

Check On Task