The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests codebase metadata (schema, routes, and source code) which constitutes untrusted external data. This creates a surface for indirect prompt injection, but the risk is mitigated as the primary purpose of the skill is local code analysis and it requires posting a Context Summary for human review.
Ingestion points: Reads structural files including db/schema.rb, config/routes.rb, Gemfile.lock, and various application layers (controllers, models, services) as defined in references/context-sources.md.
Boundary markers: The skill uses a structured 'Context Summary Template' to delimit and organize findings for the user.
Capability inventory: Utilizes file reading, directory listing, grep commands, and the get_project_context tool.
Sanitization: No explicit sanitization of the file content is mentioned before interpolation into the summary.
[DATA_EXFILTRATION]: The skill accesses sensitive application configuration and structural files such as database schemas. This is consistent with its development purpose and does not involve unauthorized network transmission or exfiltration to external domains.

load-context