The Agent Skills Directory

[PROMPT_INJECTION]: The skill instructions define an architecture for processing untrusted external source code, which inherently presents a surface for indirect prompt injection.
Ingestion points: File contents are ingested into the agent context using Read, Grep, and Glob tools as documented in references/analyze-code.md and references/quality-improve.md.
Boundary markers: The prompts for sub-agents do not explicitly include delimiters or instructions to treat source code strictly as data, which is a common practice for risk reduction.
Capability inventory: The skill has access to high-capability tools including Bash (for analysis tools), Edit (for code modification), and Task (for delegation to other agents).
Sanitization: There is no mention of input validation or content filtering for the data read from project files.
[COMMAND_EXECUTION]: The workflow documentation mentions using the Bash tool to run external analysis utilities. This is a standard and expected capability for code quality and security scanning tools to operate effectively.

code-quality-workflow