The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the Bash tool to execute a local JavaScript utility (scripts/novel-characters.mjs) and an external binary (codex) to perform text processing, report rendering, and image generation.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted text files (novels/short stories) provided by the user, which could potentially contain adversarial instructions aimed at influencing the character extraction or profiling process.
Ingestion points: The main novel text (book.txt) is read and split into overlapping chunks, which are then sequentially processed by the agent.
Boundary markers: The agent uses specific instruction templates (roster-pass.md, profile-pass.md) to guide the extraction, although the raw data chunks do not contain explicit end-of-content delimiters.
Capability inventory: The skill utilizes file system access (read/write), shell command execution, and image generation via the codex platform.
Sanitization: The report generation logic in novel-characters.mjs uses an esc() function to sanitize data before rendering HTML. The skill also includes a validate command that deterministically checks that extracted character evidence matches the original source text verbatim, mitigating the risk of model hallucinations or data poisoning during the profiling phase.

novel-characters