Agent Task Status

Verify whether named OpenClaw agents actually received formal task assignments and replied with execution status, using transcript-backed audit checks. Use w...

Security Scan

The skill's code and instructions match its stated purpose (reading local OpenClaw agent session indexes and transcripts to report assignment/report status); it requires no secrets or external installs and does not contact external endpoints.

Purpose & Capability

The name/description ask to verify assignment delivery and report status; the bundled script reads a configurable agents root (default ~/.openclaw/agents), session index files, and transcript files to extract assignments and assistant reports — exactly what the skill claims to do. No unrelated services or credentials are requested.

Instruction Scope

SKILL.md instructs running the included Python script against local session indexes and transcripts and documents flags for agent lists, base path, markers, filters, and output formats. The instructions limit actions to file reads and local output; they do not instruct network exfiltration or scanning of unrelated system paths. Note: SKILL.md and the script reference optional environment variables (OPENCLAW_*) as overrides, but these are optional and used only to change the local paths/keywords.

Install Mechanism

There is no install spec; this is an instruction-only skill plus a local Python script. No packages are downloaded or installed by the skill bundle.

Credentials

The skill declares no required environment variables or credentials. The script does read several optional env vars (OPENCLAW_AGENTS_BASE, OPENCLAW_SESSION_KEY_TEMPLATE, OPENCLAW_OUTPUT_FORMAT, etc.) as user overrides; this is proportional to its configurability. There are no requests for unrelated secrets or cloud credentials.

Persistence & Privilege

always is false and the skill does not request persistent platform privileges. The script reads and can write an output file (user-specified) but does not alter other skills' configs or system-wide settings. Autonomous invocation is allowed by default (platform normal) but the skill's actions remain file-scoped.

Assessment

This skill reads local OpenClaw agent session indexes and transcript files (default ~/.openclaw/agents). Before installing: review whether those transcripts contain sensitive data you don't want exposed; confirm the agents root (--base or OPENCLAW_AGENTS_BASE) is correct and limited to the data you intend to inspect; prefer running the script in a safe context or with a read-only copy of transcripts if you have multi-tenant concerns. The bundle includes the full Python source — if you have security concerns, scan that file for unexpected network calls or file writes (the supplied script appears to only perform local file reads and optional output-file writes). Finally, when using automation, be mindful that the agent platform may invoke the skill autonomously — limit the skill's scope via the --base flag and avoid pointing it at system-wide or other users' directories.

Use the bundled script to inspect OpenClaw session indexes and transcript files, then extract the latest assignment and structured report for each target agent.

Quick start

Run the script directly:

python3 /home/lyqadmin/.openclaw/workspace/skills/agent-task-status/scripts/check_agent_task_status.py --format summary

For more examples, read references/usage.md.

Workflow

1. Decide which agents to inspect.
   • Use --agents a,b,c for explicit targets.
   • Use --agent-file when the list comes from a file.
   • Use --discover when the deployment has many agents under the same root.
2. Set the agent storage root.
   • Default is ~/.openclaw/agents.
   • Override with --base or OPENCLAW_AGENTS_BASE in non-default environments.
3. Match the session shape.
   • Default session key template is agent:{agent}:main.
   • Override with --session-key-template if your target sessions use another pattern.
4. Match the assignment/report language.
   • Default assignment keyword is 正式任务分配：.
   • Default report prefixes are 任务： / 状态： / 结果： / 风险：.
   • Override these when the team uses different markers or another language.
5. Filter the output when needed.
   • --only-status filters by normalized status such as completed, blocked, accepted, no-assignment, assigned-no-report, error.
   • --contains filters by keyword across assignment text, parsed task, result, and risk.
6. Pick an output format.
   • table: best for human inspection
   • summary: compact overview
   • json: structured automation output
   • jsonl: line-oriented pipelines
7. Use --strict for CI/automation.
   • Exit 0: normal
   • Exit 1: partial problem such as missing assignment/report or agent error
   • Exit 2: script/runtime error
8. Use --output-file to persist results for later review or downstream automation.

Recommended commands

Human-readable table:

python3 /home/lyqadmin/.openclaw/workspace/skills/agent-task-status/scripts/check_agent_task_status.py --agents xiaocheng,xiaowen,xiaobian --format table

Only completed tasks:

python3 /home/lyqadmin/.openclaw/workspace/skills/agent-task-status/scripts/check_agent_task_status.py --discover --only-status completed --format summary

Filter by keyword:

python3 /home/lyqadmin/.openclaw/workspace/skills/agent-task-status/scripts/check_agent_task_status.py --discover --contains 自动化 --format table

Automation JSON:

python3 /home/lyqadmin/.openclaw/workspace/skills/agent-task-status/scripts/check_agent_task_status.py --agent-file ./agents.txt --format json --strict --output-file ./agent-status.json

What to inspect in the output

• sessionKey: which session was used
• sessionFile: exact transcript file path
• assignedAt / assignText: when and what was assigned
• reportAt: when the agent reported back
• task / status_raw / status_normalized / result / risk: parsed structured fields
• error: why the check failed for that agent

Limitations

• This skill assumes an OpenClaw-style agent root with sessions/sessions.json and transcript sessionFile paths.
• It only checks the target session pattern you specify; it does not automatically infer every possible routing form.
• If the assignment keyword or report field prefixes change, you must override them.
• It reports what is present in transcripts; it does not infer hidden work with no assignment/report markers.