The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides Python scripts (scripts/agent.py and scripts/process.py) that perform network operations and API interactions. scripts/agent.py communicates with CyberArk PVWA REST APIs for authentication and auditing, while scripts/process.py uses the AWS SDK to query IAM users, roles, and policies.
[INDIRECT_PROMPT_INJECTION]: The audit scripts ingest data from external sources (AWS IAM metadata and CyberArk API responses) and process it to generate JSON reports. This pattern establishes an indirect prompt injection surface where maliciously crafted metadata in a compromised cloud environment could influence the tool's findings or the subsequent analysis performed by the AI agent.
Ingestion points: AWS IAM user/role metadata (scripts/process.py) and various CyberArk API responses (scripts/agent.py).
Boundary markers: No delimiters or specific warnings for the agent to ignore embedded instructions in the ingested data were found.
Capability inventory: The skill performs network requests and writes output files to the local system (zsp_audit.json, zsp_report.json).
Sanitization: The scripts do not perform explicit validation or sanitization of the content retrieved from external APIs before writing it to the local reports.
[EXTERNAL_DOWNLOADS]: The scripts depend on standard Python packages including boto3, requests, and urllib3. These are well-known libraries used within their expected functional scope.
[PRIVILEGE_ESCALATION]: The scripts/agent.py script includes a mechanism to bypass TLS certificate verification via the SKIP_TLS_VERIFY environment variable. While documented for laboratory use, this configuration poses a risk of credential interception via man-in-the-middle attacks. Additionally, the script accepts passwords via command-line arguments, which is a sub-optimal practice that can lead to credential exposure in process logs and shell history.
[METADATA_POISONING]: There is a discrepancy between the author listed in the SKILL.md frontmatter ('mahipal') and the copyright holder specified in the LICENSE file ('mukul975'). While potentially a remnant of a template, this inconsistency represents deceptive metadata regarding the skill's origin.

implementing-zero-standing-privilege-with-cyberark