The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data which creates a surface for indirect prompt injection attacks.
Ingestion points: The skill accepts "错误信息或异常行为" (error information or abnormal behavior) as input and specifically instructs the agent to process external artifacts such as network requests, payloads, event logs, and HAR files (Phase 1).
Boundary markers: There are no explicit instructions or delimiters provided to the agent to treat this ingested data as untrusted or to ignore any potential instructions embedded within those files.
Capability inventory: The skill requires the ability to execute shell commands, HTTP scripts, CLI tools, and headless browser scripts (Phase 1).
Sanitization: The instructions do not include steps for sanitizing or validating the content of the ingested logs or traces before they are used to influence the agent's logic.
[DYNAMIC_EXECUTION]: The skill encourages the dynamic creation and execution of scripts to facilitate debugging.
Evidence: Phase 1 instructions include creating "Curl / HTTP script", "CLI invocation with a fixture", "Headless browser script", and "Throwaway harness".
Evidence: Phase 1, point 10 explicitly mentions driving a human through a shell script: scripts/hitl-loop.template.sh.
[COMMAND_EXECUTION]: The skill requires extensive use of shell and system commands to verify bug behavior and performance.
Evidence: The "Completion criterion" in Phase 1 requires the agent to provide a single command (script path, test invocation, or curl) that has already been executed.

diagnosing-bugs