The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted local data including index.html, README.md, and source files (.ts, .tsx) during the cleanup process. This creates an ingestion surface where malicious instructions embedded in a project could attempt to influence agent behavior.\n
Ingestion points: Project files such as package.json, vite.config.ts, index.html, and README.md are scanned and their content is read into the agent context.\n
Boundary markers: The instructions do not define specific delimiters or warnings to ignore instructions embedded within the files being processed.\n
Capability inventory: The skill relies on npm uninstall, npm run build, and file modification capabilities to perform its tasks.\n
Sanitization: While environment variables are redacted in terminal output via sed, the content of other project files is not explicitly sanitized before processing.\n- [COMMAND_EXECUTION]: The skill instructs the agent to execute a variety of local shell commands, including grep, sed, npm, and mktemp. These are used to audit the codebase, create temporary tracking files, remove packages, and run build verification scripts.\n- [DATA_EXFILTRATION]: The skill performs targeted scans of environment configuration files (.env, .env.local, .env.example) to identify and remove project-specific identifiers and keys. While it uses sed to redact these values from the agent's output, the agent itself must access these sensitive file paths to perform the cleanup.

lovable-cleanup