The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill context includes 144 reference files (~286,000 words) in the references/ directory. This content consists of video transcripts, scraped website data, and academic summaries.
Ingestion points: The references/ folder and its subdirectories, which the agent is instructed to read for deep color knowledge as specified in SKILL.md and references/INDEX.md.
Boundary markers: No explicit delimiters or system instructions are provided to the agent to treat this content as untrusted data or to ignore any instructions embedded within the reference text.
Capability inventory: The skill frontmatter in SKILL.md does not restrict tool access (no allowed-tools field), allowing the agent to operate with default system capabilities, including potential shell access, file system writes, and network operations.
Sanitization: There is no evidence of filtering or sanitization of the external content used as a source for these reference files to prevent the inclusion of malicious prompts.
[EXTERNAL_DOWNLOADS]: The skill documentation and instructions frequently recommend the installation of external software packages and link to various third-party tools and repositories.
Packages: The skill recommends packages such as design-book, rampensau, poline, nutelch, cusphanger, color-name-lists, apca-w3, and colorgram.
Origin: Most of the highly recommended packages are vendor-owned resources created by the skill author ('meodai'). Other references point to well-known organizations or services.

color-expert