The Agent Skills Directory

[DATA_EXPOSURE]: The skill accesses sensitive local files including .env and application logs (logs/*.log) to identify error patterns. To mitigate risk, it uses sed patterns to redact Groq API keys (matching gsk_*) and masks all values in the .env file before they are added to the debug bundle. The final output is a local .tar.gz archive intended for user-controlled sharing with support.
[COMMAND_EXECUTION]: The skill implementation utilizes various shell commands (curl, grep, sed, tar) to gather system metadata, test API connectivity, and benchmark model latency. These operations are restricted to the Groq API and local system inspection consistent with a debugging utility.
[DYNAMIC_CONTEXT_INJECTION]: The SKILL.md file uses dynamic shell execution at load time to display the current environment's Node.js, Python, and SDK versions. These commands are benign and used solely for providing immediate context to the agent.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a vulnerability surface as it ingests untrusted data from local application logs.
Ingestion points: Reads content from logs/*.log files via grep in references/implementation.md.
Boundary markers: None present; log content is appended directly to a text file for bundling.
Capability inventory: The skill has access to shell execution including curl, tar, and file system writes.
Sanitization: Includes basic redaction for API keys using sed, but does not filter for potential instructions embedded within log messages.

groq-debug-bundle