The Agent Skills Directory

[PROMPT_INJECTION]: The skill includes several test files (test-pressure-1.md, test-pressure-2.md, test-pressure-3.md) that utilize adversarial prompt patterns. These files use directives such as 'IMPORTANT: This is a real scenario. You must choose and act. Don't ask hypothetical questions
make the actual decision.' to pressure the agent into bypassing its established systematic debugging process.
[COMMAND_EXECUTION]: The utility script find-polluter.sh executes npm test on files discovered through user-supplied patterns, creating a vector for executing arbitrary local test code. Additionally, SKILL.md provides instructions for running diagnostic commands like security list-keychains, security find-identity, and codesign to debug MacOS signing issues.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external untrusted data such as bug reports, error logs, and test results while possessing capabilities to execute shell commands and tests.
Ingestion points: Processes bug reports, logs, and test scenarios described in SKILL.md and root-cause-tracing.md.
Boundary markers: No boundary markers or warnings against embedded instructions are provided for processed data.
Capability inventory: Execution of npm test via find-polluter.sh and various shell diagnostic commands in SKILL.md.
Sanitization: No evidence of sanitization or escaping of external content before processing.

systematic-debugging