The Agent Skills Directory

[PROMPT_INJECTION]: The skill's architecture is susceptible to indirect prompt injection because it processes arbitrary content from user projects using LLM-powered sub-agents.
Ingestion points: Text files discovered by scripts/scan-plaintext-files.sh and read during the review process.
Boundary markers: Absent; sub-agent instructions do not include explicit delimiters or 'ignore embedded instructions' directives for the ingested text.
Capability inventory: File system Read and Write access (for report generation) and Task management capabilities.
Sanitization: No sanitization of ingested text content is performed before processing.
[EXTERNAL_DOWNLOADS]: The testing utility tests/run-tests.sh fetches the bats-core framework from its official GitHub repository to support automated testing.

plain-language