The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection by processing external data.
Ingestion points: Project source files are scanned using rg (ripgrep) in Phase 2 of the workflow as defined in SKILL.md and references/detection-catalog.md.
Boundary markers: The assets/AUDIT-REPORT.md template uses Markdown code blocks (```kotlin) to encapsulate evidence snippets, providing structural separation but lacking explicit "ignore embedded instructions" warnings.
Capability inventory: The skill utilizes shell commands (rg, find), the android-cli for documentation lookups, and the JetBrains-focused context7 MCP.
Sanitization: The instructions do not specify sanitization or escaping of the retrieved code snippets before they are interpolated into the final audit report.
[COMMAND_EXECUTION]: The skill directs the agent to execute shell commands to perform its auditing functions.
Evidence: SKILL.md and references/discovery-playbook.md describe the use of find and rg to discover project types and search for code patterns.
Context: These tools are standard for static analysis and are used here within a defined, hardcoded set of patterns.

compose-audit