The Agent Skills Directory

[PROMPT_INJECTION]: The skill defines a workflow for agents to investigate user-provided error reports and code snippets (SKILL.md), which inherently presents a surface for indirect prompt injection. Injected instructions within user logs could attempt to influence the agent's diagnostic logic.
Ingestion points: Collection of error messages, stack traces, and reproduction steps in the 'Understand the Problem' phase (SKILL.md).
Boundary markers: The workflow does not explicitly define delimiters to isolate untrusted user data from the agent's instructions.
Capability inventory: Troubleshooting tasks including code analysis, log tracing, and the use of debugging tools.
Sanitization: The skill provides instructions for manual analysis but does not prescribe automated sanitization or validation of the input data.
[COMMAND_EXECUTION]: Educational scripts and reference materials (examples/debugging-workflow.sh, references/shell-errors.md) demonstrate the use of standard shell commands such as 'rm', 'chmod', and 'trap'. These are used within the context of demonstrating common pitfalls, cleanup operations, and proper script maintenance.
[EXTERNAL_DOWNLOADS]: Documentation and reference files (references/debugging-tools.md, references/python-errors.md) mention standard development tools and package managers, including 'pip', 'npm', 'curl', 'jq', and 'git', as part of the recommended debugging environment.

bug-detective