The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill maintains session continuity by reading and writing state information to disk (state.json and module files). While this creates an ingestion surface for potentially untrusted data from previous sessions, the skill provides robust defensive instructions to validate inputs.
Ingestion points: state.json, modules/*.md, and founders/*.md are read during the session start protocol to resume progress.
Boundary markers: The skill uses structured Markdown headers (## Raw, ## Synthesis) to separate verbatim user inputs from the agent's interpretation.
Capability inventory: The skill utilizes file read and write operations to manage the brand identity project files.
Sanitization: The skill explicitly instructs the agent to validate participant names (allowing only alphanumeric characters and hyphens) and to reject any paths containing separators (/, \, ..) to prevent path traversal attacks. It also requires outputPath to be an absolute path within the project directory.
[DATA_EXPOSURE]: The skill manages organizational brand data. It does not attempt to access sensitive system locations, environment variables, or hardcoded credentials. All disk operations are scoped to the user-provided project directory.

brand-discovery