The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process third-party hardware documentation, Device Tree Sources (DTS), and vendor SDKs to model QEMU boards and SoCs. This establishes an attack surface where malicious or malformed instructions embedded in these external specifications could influence the agent's behavior during the code generation process.\n
Ingestion points: Hardware manuals, vendor documentation, DTS files, and firmware images referenced in SKILL.md and assets/board-contract-template.md.\n
Boundary markers: The skill mandates that all external findings and modeling decisions be recorded in a project-local audit.md file within a dedicated .oh-my-qemu/ task directory, providing a structural audit trail, though it does not define specific LLM-targeted delimiters for the data itself.\n
Capability inventory: The agent is instructed to modify QEMU source code in hw/, create and register qtests in tests/qtest/, and execute build and test commands using meson as described in SKILL.md and assets/board-contract-template.md.\n
Sanitization: No specific text sanitization or escaping mechanisms are defined; the skill relies on the agent's technical analysis and the provided technical templates to ensure code integrity.\n- [EXTERNAL_DOWNLOADS]: The skill's documentation regarding boot path verification mentions the acquisition of external assets such as kernel images or firmware.\n
Evidence: In references/boot-paths.md, the skill provides security guidelines for functional tests that download assets, explicitly recommending that these assets be served locally and verified using pinned hashes to ensure integrity.

qemu-board-modeling