The Agent Skills Directory

[COMMAND_EXECUTION]: The find-polluter.sh script automates the execution of local tests using npm test to identify tests that create side effects (pollution). This is a standard development utility used to isolate problematic test code.
[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to ingest and analyze external data like error messages, stack traces, and log outputs. This represents a standard surface for processing technical data during debugging tasks. The instructions emphasize systematic analysis rather than blind execution of embedded content.
[DATA_EXPOSURE]: Documentation includes examples for verifying the presence of environment variables (secrets) at component boundaries to debug configuration propagation issues. These are diagnostic checks (e.g., checking if a variable is SET or UNSET) and do not involve sending data to external services.
[PROMPT_INJECTION]: The skill package includes scenario-based tests (test-pressure-*.md) designed to evaluate the AI's adherence to its core debugging principles under simulated stress. These are internal evaluation tools for the skill itself and do not represent malicious attempts to bypass platform safety guidelines.

systematic-debugging