The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/collect_summary.py executes local dce CLI commands using subprocess.run to gather operational data. The implementation passes arguments as a list, which prevents shell injection vulnerabilities and is a standard method for interacting with vendor-provided platform tools.
[PROMPT_INJECTION]: The skill processes JSON data from external command outputs, creating an indirect prompt injection surface. However, the risk is mitigated by strict instructional constraints in SKILL.md regarding data integrity and the use of a Python intermediary to sanitize and structure the data.
Ingestion points: Command output from the dce CLI processed in scripts/collect_summary.py.
Boundary markers: The agent is constrained by a fixed Markdown response template in SKILL.md.
Capability inventory: Subprocess calls to the dce tool in scripts/collect_summary.py; no generic shell access or network write capabilities are present.
Sanitization: The Python script scripts/collect_summary.py parses raw output and emits structured NDJSON, filtering out raw secrets before the agent receives the data.
[DATA_EXFILTRATION]: While the skill retrieves sensitive operational metrics, including API key metadata, it includes explicit instructions in SKILL.md to 'Never expose raw API keys, tokens, credentials, or secrets.' The script performs internal aggregation and reports only counts and statuses, ensuring that sensitive values are not transmitted or displayed.

ai-ops-daily-summary