The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes output from external firmware images, specifically capturing boot logs and serial terminal data. This constitutes a vulnerability surface where a maliciously crafted firmware image could produce output intended to override agent instructions.
Ingestion points: The agent is instructed to read and verify boot-log.txt and live serial output from QEMU processes.
Boundary markers: The instructions lack specific delimiters or warnings for the agent to disregard potentially malicious commands embedded within the log output, relying instead on simple pattern matching for boot success.
Capability inventory: The skill environment permits shell command execution (qemu-system-*, docker, gunzip) and file system access.
Sanitization: No sanitization or filtering of terminal output is specified before the content is processed by the agent.
[COMMAND_EXECUTION]: The skill provides templates for executing complex shell commands to facilitate emulation.
Evidence: Instructions include the use of qemu-system-aarch64 with specific hardware mapping parameters and docker run to provide isolated environments. These commands are legitimate and proportional to the skill's stated purpose.
[EXTERNAL_DOWNLOADS]: The skill references several external sites for downloading firmware and tools.
Evidence: It points to official domains such as downloads.openwrt.org and qemu.org, as well as official GitHub repositories for tianocore/edk2 and espressif/qemu. These are well-known and trusted sources within the embedded development community.
[CREDENTIALS_SAFE]: The skill documentation includes explicit safety guidance regarding secret management.
Evidence: references/host-mocking.md contains strict rules forbidding the inclusion of real keys, passwords, or serial numbers in test fixtures, mandating the use of explicitly constructed mock data instead.

fw-emulation