The Agent Skills Directory

[DATA_EXFILTRATION]: The skill accesses the local file path ~/.claude/company-context.md, which is used to store highly sensitive company information.
The file contents include absolute revenue figures, burn rates, specific runway periods, and lists of customer and employee names.
Although the skill instructions mandate an anonymization protocol, it explicitly prepares this sensitive data for use in external tool calls, such as web searches and API requests.
[INDIRECT_PROMPT_INJECTION]: The skill acts as a data ingestion layer that loads external content into the agent's active memory.
Ingestion points: The agent reads the file ~/.claude/company-context.md at the start of every session.
Boundary markers: The instructions do not specify the use of delimiters or instructions to ignore potential commands embedded within the context file.
Capability inventory: The skill is designed to interact with external APIs and web search tools using the loaded context.
Sanitization: The skill provides a detailed anonymization-protocol.md to filter data before external transmission, but this serves as a guideline for the LLM rather than a hard programmatic filter, leaving a risk of accidental or coerced data exposure.

context-engine