The Agent Skills Directory

[Unverifiable Remote/Local Code] (MEDIUM): The skill relies on local Python scripts (scripts/api_scaffolder.py, scripts/database_migration_tool.py, scripts/api_load_tester.py) that are not included in the provided files. Their behavior cannot be audited, posing a risk of arbitrary command execution if those scripts contain malicious logic.
[Indirect Prompt Injection Surface] (HIGH): The skill is designed to ingest and process untrusted external data (project paths, backend code, and database schemas).
Ingestion points: User-specified <project-path> and <target-path> in SKILL.md.
Boundary markers: No explicit delimiters or instructions to ignore embedded commands in the processed data are mentioned.
Capability inventory: The skill allows the agent to execute shell commands, run Python scripts, perform npm operations, and deploy via kubectl and docker.
Sanitization: No sanitization or validation of the processed code is described.
Risk: An attacker could place malicious instructions inside a target codebase's comments or configuration files, which the agent might execute with high privileges during 'analysis' or 'optimization' tasks.
[Privilege Escalation Risk] (MEDIUM): The inclusion of deployment commands (kubectl apply, docker-compose up) suggests the agent may have access to production or staging environments. Executing these commands based on analysis of untrusted local files is a high-risk pattern.

senior-backend