The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to analyze external, potentially untrusted repositories, which creates a surface for indirect prompt injection attacks where malicious files could attempt to influence the agent's behavior.
Ingestion points: The scripts/detect-project.py script crawls and reads content from various files in the target repository, including manifests, configuration files, and source code.
Boundary markers: The agent is provided with clear instructions in SKILL.md to treat inferred commands as candidates and to verify results. The tool output is structured in a machine-readable JSON format, which helps separate data from instructions.
Capability inventory: The analysis script is restricted to local read-only operations on the filesystem and prints its results to standard output. There are no network operations, file-writing tasks, or dynamic code execution capabilities (such as eval or exec) present in the skill's code.
Sanitization: The script implements several safety measures, including a 1MB file size limit for reading text, using json.dumps for output formatting, and explicitly excluding the values of environment variables from its reports.

analyze-deployment