The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The scripts/collect-context.sh script captures user-provided data, such as the command line used to trigger a failure, and interpolates it directly into a Markdown report without sanitization. If this report is subsequently processed by the agent, it could lead to indirect prompt injection.\n
Ingestion points: CLI arguments --command, --qemu, and --log in scripts/collect-context.sh are directly included in the output.\n
Boundary markers: The generated report (audit.md) lacks specific delimiters or warnings to prevent the agent from interpreting injected text as instructions.\n
Capability inventory: scripts/collect-context.sh performs file writes; SKILL.md directs the use of gdb, lldb, ninja, meson, and git via shell commands.\n
Sanitization: There is no sanitization or escaping performed on the captured command_line or path variables before they are written to the report file.\n- [COMMAND_EXECUTION]: The skill requires the agent to execute a local shell script and several host-level debugging and build tools. While appropriate for the task of QEMU debugging, this involves significant interaction with the host system's command line.

qemu-debug