The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it analyzes untrusted source code from the repository. Maliciously crafted comments or strings in the audited files could potentially manipulate the inspector subagent's findings or behavior.
Ingestion points: Source code files from user-defined areas are read and processed using LSP tools such as mcp__lsp__open_document and mcp__lsp__get_references, as defined in the SKILL.md and references/check-taxonomy.md files.
Boundary markers: The instructions do not provide delimiters or clear directives for the agent to treat the audited code as untrusted data or to ignore any natural language instructions embedded within that code.
Capability inventory: The skill has access to multiple LSP tools for deep code inspection, launches a subagent via Agent(subagent_type=inspector), and can write files to the disk to persist reports.
Sanitization: There is no evidence of filtering or sanitization of the ingested code content before it is processed by the agent.
[COMMAND_EXECUTION]: The skill instructions direct the agent to execute a shell command (touch /tmp/.inspector-lsp-global-ready) to create a coordination flag in a global temporary directory for the background auditing process.

inspect