The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the Bash tool to execute grep and a specific Python script located at ~/.claude/scripts/scan-negative-framing.py. These commands act as a pre-filter to detect negative framing patterns in target files before performing semantic analysis.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes untrusted data from user-provided files to perform tone and framing analysis. This creates a surface for indirect prompt injection where malicious instructions embedded in the analyzed text could attempt to manipulate the agent's scoring or the reframed output, especially when the --fix capability is enabled.
Ingestion points: Target files are read during the analysis phase (SKILL.md).
Boundary markers: Instructions require the agent to skip frontmatter and code blocks, providing some separation of data.
Capability inventory: The skill is granted Bash, Write, and Edit tools, enabling it to execute commands and modify files in place.
Sanitization: There is no explicit sanitization or instruction-filtering for the content being analyzed before it is passed to the LLM for evaluation against the provided rubrics.

joy-check