The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to ingest and analyze test failure output and source code files, which creates a potential surface for indirect prompt injection if those files contain adversarial instructions.
Ingestion points: The agent reads command output from make test and pytest, as well as file contents via the git diff and Edit tools in the SKILL.md file.
Capability inventory: The skill allows the agent to execute shell commands (tests), read files, and modify code, providing a platform for actions if an injection occurs.
Boundary markers: The instructions do not specify the use of delimiters or specific warnings to ignore instructions embedded within the test data or code files.
Sanitization: No explicit sanitization or validation logic is defined for the external content processed during the fixing workflow.

test-fixing