The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted workspace data (code, configuration, and metadata) during the validation inventory and catalog check steps (SKILL.md steps 3-4). This ingestion point creates an attack surface where malicious code comments or documentation could attempt to influence the agent's behavior.
Ingestion points: reads all files in the cloned workspace including processes/, modules/, team.json, and i18n/.
Boundary markers: No specific delimiters or "ignore" instructions are mentioned for the content being read.
Capability inventory: Includes pnpm install (vendor CLI), fcode clone, fcode pull, and writing the APP_VALIDATION_REPORT.md file.
Sanitization: No explicit sanitization of audited content is described.
[EXTERNAL_DOWNLOADS]: The skill fetches code from the Factorial Code platform using fcode clone and fcode pull. It also instructs the agent to perform a global installation of the vendor's CLI tool (@factorialco/fcode-cli) if it is not already present. As these resources belong to the skill author's organization, this represents standard vendor functionality.
[COMMAND_EXECUTION]: The skill utilizes several shell commands to manage the workspace and perform checks, such as fcode status, fcode variables:status, and fcode team:status. These are limited to the vendor's toolset and are consistent with the skill's purpose.

fcode-code-validation