The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted source code and PR diffs to perform its primary function, creating an attack surface for indirect prompt injection.
Ingestion points: The skill audits files and PR diffs provided by the user in the repository.
Boundary markers: The instructions do not prescribe the use of specific delimiters or 'ignore' instructions to prevent the agent from obeying commands embedded in audited code.
Capability inventory: The agent may have capabilities such as file modification or shell access that could be targeted by malicious payloads in audited files.
Sanitization: No sanitization of the audited content is performed.
[COMMAND_EXECUTION]: The skill's rules provide numerous shell commands for pattern detection.
Evidence: Rule files contain snippets using grep, find, and git to analyze local source code. These are standard tools for code auditing and perform local operations.
[DYNAMIC_EXECUTION]: A dependency audit rule involves running a Node.js script.
Evidence: rules/over-eng-dependency-creep.md includes a node -e command that executes a hardcoded script to verify package.json contents. The script is benign and confined to local analysis.
[EXTERNAL_DOWNLOADS]: The skill references several external resources for its methodology.
Evidence: The skill links to research papers on arxiv.org and community tools on GitHub (stop-slop, AI-SLOP-Detector). These are used for documentation and are not executed automatically by the skill.

code-slop