The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains thousands of lines of repetitive filler text across eight reference files, which acts as a form of context stuffing or obfuscation.
Files such as references/architecture-patterns.md, references/security-best-practices.md, and references/performance-optimization.md each contain 150 identical or near-identical sections of boilerplate regarding kernel interactions, eBPF, and Zero Trust architectures.
This technique can be used to exhaust an AI agent's context window or obscure malicious patterns from human review.
[PROMPT_INJECTION]: The skill presents an attack surface for indirect prompt injection through the processing of untrusted external data.
Ingestion points: The agent is instructed to read and process schemas from user-provided files and external Schema Registry subjects (SKILL.md).
Boundary markers: The instructions lack clear delimiters or specific directives for the agent to treat schema content as data rather than instructions.
Capability inventory: The skill utilizes network capabilities via curl and system capabilities for Docker deployment and CI/CD script execution (SKILL.md).
Sanitization: No sanitization or validation logic is defined to inspect external schemas for embedded malicious instructions.
[EXTERNAL_DOWNLOADS]: The skill references downloads and dependencies from established and reputable sources.
Documentation includes pulling official Docker images from Confluent (confluentinc/cp-schema-registry).
CI/CD workflow examples reference standard Python libraries like requests, avro, and fastavro from official registries.
These references are documented neutrally and are consistent with the skill's primary purpose of schema management.

data-schema-registry