The Agent Skills Directory

[DYNAMIC_CONTEXT_INJECTION]: The skill uses dynamic context placeholders to perform environment discovery during the skill loading process.
Evidence: The code snippet !command -v foremost... in SKILL.md checks for the installation of the binary and retrieves its version string.
Evidence: The code snippet !test -f scripts/foremost/recover-deleted-files.sh... in SKILL.md verifies the existence of local wrapper scripts.
These commands are used for status reporting and do not involve sensitive data access, user input injection, or network exfiltration.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted binary data from external disk images, which presents a surface for indirect prompt injection if recovered data is interpreted by the agent.
Ingestion points: Disk images and raw partitions accessed via the -i flag as documented in SKILL.md.
Boundary markers: Absent; there are no instructions to the agent to treat recovered file content as untrusted or to ignore embedded instructions.
Capability inventory: The skill facilitates shell command execution for forensic tools and local bash scripts.
Sanitization: No sanitization of the recovered files or audit logs is performed before they are accessed by the agent.
This risk is inherent to the forensic analysis use case where the data being analyzed is by definition unknown and potentially adversarial.

foremost