The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Fetches service status and validates API credentials using official Adobe domains including status.adobe.com, adobelogin.com, and adobe.io. These are established service endpoints necessary for the skill's primary function.
[REMOTE_CODE_EXECUTION]: An automated scan identified a command piping a URL to Python. Detailed review shows the command curl -sf https://your-app.com/health | python3 -m json.tool is used. This utilizes the Python standard library's JSON formatting module to pretty-print health check results and does not execute remote code.
[COMMAND_EXECUTION]: Uses standard administrative tools including kubectl for cluster logs, curl for API testing, and gcloud/aws for secret management. These commands are properly scoped to the operational tasks described in the runbook.
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface where it processes potentially untrusted data from application logs.
- Ingestion points: Ingests external data through kubectl logs commands in SKILL.md.
- Boundary markers: Absent; the skill pipes raw log output directly to grep for analysis.
- Capability inventory: Includes subprocess execution via kubectl, curl, and cloud provider CLI tools.
- Sanitization: Uses grep and tail to filter for specific error patterns, which provides basic structural filtering but does not fully sanitize log content before it enters the agent's context.

adobe-incident-runbook