The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local shell commands and a bundled Node.js script (scan.mjs) to perform text analysis. It uses standard utilities like git, mkdir, and cp for version control checks and creating file snapshots in a temporary scratch directory. These operations are limited to the project files and are necessary for the skill's stated purpose of auditable text editing.
[DATA_EXPOSURE]: The skill reads and modifies text files provided by the user. It creates backup copies of files in a <scratch>/slop/ directory before performing edits. This is a standard safety measure for text transformation tools and does not involve accessing sensitive system files or environment variables.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface because it processes untrusted user-provided text. A malicious file could contain instructions designed to influence the agent during the 'Fix' phase. However, the skill includes mitigation strategies such as masking code blocks/quotes during scanning and providing the agent with a strict 'Fix Playbook' that mandates minimum effective edits and forbids inventing new facts, which limits the potential impact of such injections.

remove-ai-slop