The Agent Skills Directory

[DATA_EXFILTRATION]: The skill explicitly requests and reads highly sensitive business documents, including 5-10 recently signed agreements and internal escalation matrices. While this is the intended functionality to establish a legal practice profile, legal contracts contain sensitive commercial data, trade secrets, and PII.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data (signed contracts) to extract "actual review guideline positions" which are then written into a persistent configuration file. This creates a vulnerability where instructions embedded within the legal documents could influence the agent's logic or the resulting profile.
Ingestion points: User-provided seed files (signed agreements) and escalation matrices.
Boundary markers: None defined; the skill lacks instructions to use delimiters or ignore embedded directives when processing the agreements.
Capability inventory: The skill has the capability to read and write local configuration files (~/.claude/plugins/config/...) which govern the behavior of other skills in the plugin.
Sanitization: There is no mention of sanitizing, escaping, or validating the content of the ingested files before they are used to generate the final CLAUDE.md profile.

cold-start-interview