The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The methodology involves the agent ingesting and analyzing untrusted external data such as application logs and user-provided symptom reports, which creates a potential surface for indirect prompt injection. \n
Ingestion points: Project source code, application log files (e.g., ~/.qwen/logs), and external user symptom reports described in SKILL.md and examples/headless-bg-agent-empty-stdout.md. \n
Boundary markers: The skill does not prescribe specific delimiters or instruction-ignore blocks when the agent parses or summarizes these external data sources. \n
Capability inventory: The skill requires the agent to perform file system reads (cat), write code instrumentation (console.log), and create persistent investigation journals (file writes). \n
Sanitization: No explicit sanitization or input validation steps are provided for handling data retrieved from external logs. \n- [SAFE]: The skill is purely instructional and documents standard industry best practices for software troubleshooting. All referenced resources, such as the qwen-code repository and ~/.qwen/logs directory, are consistent with the official tools and infrastructure of the author, qwenlm.

structured-debugging