The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains explicit instructions to override standard security evaluation logic. In references/security.md and references/code-reviewer.md, agents are directed to disregard findings related to missing authentication or authorization guards, which could lead to the oversight of critical security flaws.\n- [DATA_EXFILTRATION]: The system analysis component is configured to access and process sensitive environment files, which typically contain production credentials and secrets.\n
Evidence: references/system-analyzer.md specifies reading .env.local, .env.sit, .env.uat, and .env.prod. references/system-analyzer-cli-tools.md includes commands to export variables from these files for database access.\n- [COMMAND_EXECUTION]: The skill makes extensive use of the Bash tool to execute high-privilege system commands for infrastructure management and database interaction.\n
Evidence: references/system-analyzer-cli-tools.md provides templates for kubectl, psql, argocd, and docker commands.\n- [PROMPT_INJECTION]: The orchestration logic in SKILL.md creates an indirect prompt injection surface by interpolating the contents of project files and previous agent outputs directly into the prompts of specialists with shell access.\n
Ingestion points: CLAUDE.md, AGENTS.md, and .env files.\n
Boundary markers: Lacks robust 'ignore instructions' warnings or escaping for interpolated content.\n
Capability inventory: Specialist agents have access to Bash, Edit, and Write tools.\n
Sanitization: No sanitization or validation of external project data is performed before prompt interpolation.

neo-team-claude