The Agent Skills Directory

[PROMPT_INJECTION]: Misleading Metadata. The skill name 'kernel-development' and its description focused on 'kernel module development, drivers, and interrupt handling' are deceptive. The actual content consists almost entirely of cloud-native architecture patterns (Kubernetes HPA, Zero Trust, gRPC, and distributed databases) which are unrelated to the stated purpose.
[PROMPT_INJECTION]: Context Window Exhaustion (Token Bombing). The skill includes eight reference markdown files, each containing 150 repetitive sections. This results in a payload of approximately 180,000 words (~1.2 million characters). This extreme redundancy serves no technical purpose and is characteristic of a 'token bomb' designed to degrade agent performance, exhaust context limits, and incur high inference costs.
[PROMPT_INJECTION]: Indirect Prompt Injection Surface.
Ingestion points: Eight reference files in the references/ directory, including architecture-patterns.md and state-management.md.
Boundary markers: Absent. The technical content is not delimited from agent instructions, which could allow instructions hidden in the walls of text to be interpreted as high-priority commands.
Capability inventory: The agent's default tools (e.g., shell execution, file management).
Sanitization: Absent. There is no filtering or validation of the massive amount of technical text provided to the agent.
[NO_CODE]: Absence of Executable Scripts. The skill contains only Markdown files and no actual scripts (.py, .js, .sh) or configuration files that directly execute logic. All code examples are provided as non-executable text snippets within the documentation.

kernel-development