The Agent Skills Directory

[PROMPT_INJECTION]: The skill is designed to autonomously ingest and process varied data from the current project, including codebase logic, transaction sales logs, and pricing research studies. It lacks explicit boundary markers or sanitization procedures for this ingested content, creating an attack surface for indirect prompt injection if malicious instructions are embedded within the analyzed project files. Under adversarial conditions, this could lead the agent to follow instructions contained in data rather than the skill's own logic.
Ingestion points: Reads project codebase, transaction data, demographics, and competitive price tracking data.
Boundary markers: Absent; the instructions do not specify delimiters for external data.
Capability inventory: File system read and write access (including docs/ and ~/.claude/projects/).
Sanitization: Absent; no escaping or validation of project-supplied strings is defined.
[DATA_EXFILTRATION]: The skill attempts to locate and write execution metadata to ~/.claude/projects/skill-telemetry.md. While this interaction appears intended for local telemetry and project memory within the specific platform's environment, writing to hidden directories in the user's home folder (~/) represents an out-of-scope file operation that interacts with sensitive directory structures.

pricing-sensitivity