The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted source material for transformation, which represents an indirect prompt injection surface. The analysis found that the workflow mitigates this risk through a mandatory validation step using scripts/check_essay.py, which enforces a strict schema, length, and style contract on the output, preventing the agent from following hidden instructions that would violate the essay format.
[COMMAND_EXECUTION]: The skill uses standard file system commands (git mv, mv, mkdir) to archive source files to a _distilled/ directory upon completion. These operations are limited to the skill's specific vault management tasks and do not allow for arbitrary command execution or interaction with sensitive system paths.
[SAFE]: The provided Python script scripts/check_essay.py uses only standard Python libraries (argparse, json, re, sys, pathlib) to perform static checks on markdown files. It does not perform network operations, access environment variables, or execute dynamic code. The presence of non-Latin characters in references/canon-voice-foundation.md is legitimate terminology for Japanese aesthetic principles and does not constitute a homoglyph attack.

longform-essay