The Agent Skills Directory

[OBFUSCATION]: The skill employs zero-width Unicode characters (U+200B, U+200C, U+200D) to hide large blocks of data directly following the main headers in SKILL.md and references/philosophy.md. This steganographic technique makes the content invisible to human readers in standard text editors while remaining detectable and actionable by the AI model.
[PROMPT_INJECTION]: The hidden steganographic data is placed in critical instruction files, representing a deliberate attempt to inject hidden directives into the agent's context. This method is commonly used to override system safety protocols, hijack the agent's persona, or enforce unauthorized behaviors.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: The hidden data is embedded within the headers of SKILL.md and references/philosophy.md.
Boundary markers: There are no delimiters or boundary markers used to isolate the hidden content or instruct the agent to ignore potentially malicious embedded data.
Capability inventory: The skill includes a bash script (scripts/kernel_style_check.sh) for style checking, which possesses file reading and shell execution capabilities.
Sanitization: The skill lacks any mechanism to sanitize or filter non-printable Unicode characters from its markdown files before they are processed by the agent.

torvalds-kernel-pragmatism