The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes shell commands defined in local configuration files (docs/agents/stack.md). These include commands for starting services, running development servers, and building binaries.
[PROMPT_INJECTION]: The skill processes untrusted input from multiple sources to derive QA steps and implement fixes, which presents a surface for indirect prompt injection.
Ingestion points: Reads configuration from docs/agents/issue-tracker.md, docs/agents/stack.md, and quality-gates.md. It also ingests external data from ticket comments and codebase changes via git diff.
Boundary markers: No explicit delimiters or instructions are used to separate trusted skill instructions from untrusted content retrieved from the issue tracker or code repository.
Capability inventory: The agent can execute shell commands (commands.dev, commands.build), perform network requests using curl for API testing, and modify source code files to fix detected issues.
Sanitization: There is no evidence of sanitization or validation of the content retrieved from external sources or files before it is used to generate commands or fixes.

qa-local