The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external, untrusted data including logs, traces, and user reports, creating a vulnerability surface where malicious instructions could be embedded in the data being analyzed.
Ingestion points: SKILL.md (capture logs, traces, environment details) and resources/implementation-playbook.md (Gather Information phase, User reports).
Boundary markers: Absent. The instructions do not specify the use of delimiters or 'ignore embedded instructions' warnings when processing external data.
Capability inventory: resources/implementation-playbook.md utilizes debuggers (pdb, dlv), version control tools (git bisect), and memory analysis tools (v8.writeHeapSnapshot()).
Sanitization: Absent. There are no descriptions of data sanitization, filtering, or escaping before the external content is processed by the agent.
[DYNAMIC_EXECUTION]: The skill employs dynamic execution and analysis techniques that inspect or modify runtime state, which is standard for debugging but presents a functional risk if misused.
Evidence: Documentation in resources/implementation-playbook.md for using JavaScript debugger; statements, Python breakpoint(), and Node.js v8.writeHeapSnapshot() to capture process memory.

debugging-strategies