The Agent Skills Directory

[COMMAND_EXECUTION]: The skill documentation and several internal scripts (e.g., bulk-preprocess.ts, ea-parse.ts, kb-generate-index.ts) require the execution of Deno commands with significant permissions (-A or specific read/write flags). These commands are intended to manage a local ebook collection and SQLite database (metadata.db), which is a standard pattern for developer-oriented tools but involves broad filesystem access.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to ingest and process untrusted external text from ebooks in .txt or .epub formats. Maliciously crafted instructions inside an ebook could potentially influence the LLM's output during the concept and entity extraction phases. While the extraction prompts in scripts/ea-extract.ts use clear headers (e.g., ### Text to Analyze:) to isolate book content, they do not include explicit 'ignore embedded instructions' guardrails. Evidence of this ingestion surface is found in scripts/ea-extract.ts where chunked text is interpolated directly into prompts.

ebook-analysis