The Agent Skills Directory

Command Execution: The skill instructs the agent to execute various CLI tools, including codex-security, endorctl, snyk, and semgrep. These operations are fundamental to the skill's purpose of acquiring and publishing security scanner findings.
Data Ingestion (Indirect Prompt Injection Surface): The skill processes security reports in JSON and CSV formats from external vendors. This ingestion point represents a potential surface for indirect prompt injection if source data were specifically crafted to influence agent behavior. However, the skill mitigates this through a mandatory preview and confirmation workflow.
Ingestion points: Security report files (JSON, CSV, JSONL) processed via the codex-security publish findings command (SKILL.md).
Boundary markers: The skill requires the use of --dry-run to preview findings and explicit user confirmation (--yes) before final import or publication.
Capability inventory: The skill utilizes subprocess execution for scanner CLI tools and performs network operations when publishing findings to the Cloud destination.
Sanitization: The skill relies on the internal parsing and validation logic of the codex-security CLI tool to handle vendor-specific data structures safely.
Environment Configuration: The skill uses environment variables like CODEX_SECURITY_ENABLE_EXTERNAL_IMPORTS as a safety gate and CODEX_SECURITY_CLOUD_BASE_URL to configure API endpoints, which is a standard approach for managing tool behavior across different environments.

publish-findings