The Agent Skills Directory

[COMMAND_EXECUTION]: The skill manages long-running analysis tasks by spawning background worker processes via subprocess.Popen in scripts/cli/worker_utils.py. The command is restricted to re-executing the current Python interpreter with the skill's own parameters.
[COMMAND_EXECUTION]: The utility in scripts/cli/notify.py executes local system binaries (openclaw or clawdbot) via subprocess.run to send progress updates to external agent platforms.
[DATA_EXFILTRATION]: Task updates and session metadata can be sent to a user-configured external endpoint via the ASYNC_TASK_PUSH_URL environment variable in scripts/cli/notify.py. This is an intended integration feature.
[PROMPT_INJECTION]: The skill ingests data from external databases and files, representing a surface for indirect prompt injection.
Ingestion points: External database query results and user-uploaded file contents in scripts/cli/cmd_db.py and scripts/cli/cmd_file.py.
Boundary markers: Lacks explicit delimiters for untrusted data interpolated into internal prompts.
Capability inventory: Subprocess execution in scripts/cli/worker_utils.py and scripts/cli/notify.py, and external network requests in scripts/data_agent/client.py and scripts/cli/notify.py.
Sanitization: Relies on structured API schemas but does not implement explicit content sanitization before processing data via the agent reasoning loop.

alibabacloud-data-agent-skill