The Agent Skills Directory

[COMMAND_EXECUTION]: The skill involves the execution of local Deno scripts (scripts/voice-check.ts and scripts/dialogue-audit.ts) to analyze dialogue patterns. These scripts are run with the --allow-read flag, which restricts their capability to reading the input text files provided by the user.
[PROMPT_INJECTION]: The skill is designed to process untrusted text (fiction drafts) provided by the user, which presents a surface for indirect prompt injection.
Ingestion points: The analysis scripts read content from local files or text strings provided via command-line arguments.
Boundary markers: No explicit delimiters or boundary markers are utilized within the scripts to distinguish between data and potential instructions.
Capability inventory: The processing scripts have read access to the local filesystem; the agent is instructed to write summary reports to the project directory for persistence.
Sanitization: No sanitization is performed on the input text; however, the scripts only perform regex-based linguistic analysis and do not execute the content of the ingested data.
[SAFE]: No malicious patterns, obfuscation, or unauthorized data exfiltration behaviors were detected. All external operations are local and consistent with the skill's stated purpose of dialogue diagnosis.

dialogue