The Agent Skills Directory

[DATA_EXFILTRATION]: The skill directs the agent to locate and read environment configuration files such as .env.example and .env.local.example to extract company metadata. Accessing environment variable files, even example variants, poses a high risk of sensitive data exposure, as these files often contain or are located near actual secrets. Additionally, the 'Data Collection Discovery' phase (Phase 1.1) instructs the agent to search the codebase for sensitive authentication patterns including token, bearer, authorization, and sk-. This behavior acts as a discovery and harvesting mechanism for credentials, bringing potentially sensitive keys into the agent's operational context.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a significant attack surface by ingesting untrusted data from the codebase, README files, and marketing sites to generate legally binding documents. Malicious instructions placed in these audited files could influence the agent to generate weakened liability protections or unauthorized clauses.
Ingestion points: Project metadata (package.json), documentation (README.md), environment templates (.env.example), and marketing site content.
Capability inventory: The agent drafts binding legal agreements, determines governing law, and defines liability limitations across multiple scripts.
Sanitization: There are no explicit instructions provided to validate or sanitize the audited content before it is used to populate legal templates.
Boundary markers: The skill lacks explicit boundary markers or instructions to ignore embedded commands within the files it audits.

legal-tos-privacy