The Agent Skills Directory

[PROMPT_INJECTION]: The skill creates an attack surface for indirect prompt injection by ingesting untrusted runtime data during the debugging process.
Ingestion points: The debug_read tool in SKILL.md retrieves execution logs from a local capture server.
Boundary markers: The skill instructions do not prescribe the use of delimiters or safety warnings to distinguish data from instructions when analyzing logs.
Capability inventory: The skill possesses capabilities to modify files, start a local server, and read execution state using the debug_start and debug_read tools.
Sanitization: No data validation or sanitization of the captured variable values or execution state is performed before analysis.
[COMMAND_EXECUTION]: The skill facilitates dynamic code modification by instructing the agent to insert instrumentation snippets into the user's project.
Evidence: The workflow and instrumentation_examples in SKILL.md guide the agent to insert fetch() calls targeting http://localhost:PORT/log into the application's source code to monitor variable values and errors.

debug