The Agent Skills Directory

[COMMAND_EXECUTION]: The validation scripts use unsafe methods to execute commands defined in the target codebase's configuration files.
Evidence: scripts/validate.sh uses eval "$cmd" to execute strings parsed from package.json.
Evidence: scripts/validate.ps1 uses Invoke-Expression $Command to execute strings parsed from package.json.
[PROMPT_INJECTION]: The skill processes untrusted external data (feature specifications and project configurations) which can influence the agent's behavior or trigger unintended actions.
Ingestion points: Specification files located at specs/{feature-name}.md and project manifest files like package.json.
Boundary markers: None; the skill reads and interprets the specification content as a 'single source of truth' without delimiters.
Capability inventory: Access to system shell (bash, powershell), package managers (npm, npx, go, cargo), and file system write operations.
Sanitization: No sanitization or validation is performed on the command strings extracted from project configuration files before execution.

apex-replatform