The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides numerous patterns for tokenizing, tagging, and translating natural language text from external sources (e.g., message inboxes) without including instructions or code for sanitizing input or using boundary markers. This creates an attack surface where malicious instructions embedded in the input text could influence the agent's behavior during analysis or subsequent processing.
Ingestion points: The text parameter in functions such as tokenizeWords, tagPartsOfSpeech, sentimentScore, and classifyText in both references/nlp-patterns.md and references/translation-patterns.md.
Boundary markers: Absent; the skill does not include delimiters or instructions to treat input as data only.
Capability inventory: Framework capabilities include text analysis (named entity recognition, sentiment scoring) and translation.
Sanitization: Absent; no escaping or filtering logic is provided for the input strings.
[METADATA_POISONING]: The skill documentation references fictional future versions of software (e.g., Swift 6.3, iOS 26) and provides links to an unofficial third-party domain (sosumi.ai) for framework documentation instead of official resources. These misleading references could cause an agent to provide hallucinated technical details or rely on unverified external documentation.

natural-language