The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external untrusted data such as error messages, stack traces, and diagnostic logs to perform root cause analysis.
Ingestion points: The agent is instructed to read stack traces and logs to identify bugs, as detailed in SKILL.md Phase 1 and references/root-cause-tracing.md.
Boundary markers: There are no explicit instructions to use delimiters or ignore embedded instructions within these data sources to prevent the agent from obeying instructions hidden in logs.
Capability inventory: The skill has the ability to execute shell commands (e.g., npm test, find) and is intended to facilitate code modifications to fix bugs.
Sanitization: No sanitization, escaping, or filtering of external input is mentioned or performed before the agent analyzes the content.
[COMMAND_EXECUTION]: The skill provides a shell script (scripts/find-polluter.sh) and diagnostic bash examples that execute local commands like find, env, and npm test. These are standard development tools for the intended debugging use case. The scripts/find-polluter.sh script suppresses test output (> /dev/null 2>&1), which reduces user visibility into the individual test executions during the bisection process.

systematic-debugging