The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted filenames from the workspace and interpolates them into a shell command using substitution, creating an injection surface.
Ingestion points: Local filenames and directory paths discovered in the codebase via the find command in SKILL.md.
Boundary markers: None present. There are no instructions for the agent to sanitize or escape filename inputs.
Capability inventory: The skill possesses the capability to execute shell commands such as pwd, find, wc, tree, and sort.
Sanitization: Absent. Step 3 uses the pattern wc -l $(find ...) which allows the shell to evaluate and execute any shell metacharacters or subcommands contained in a filename (e.g., a file named `id`.py or $(reboot).rs).
[COMMAND_EXECUTION]: The skill utilizes several shell commands to extract metadata and analyze the codebase organization.
Evidence: Shell commands and pipelines are used across all steps: pwd (Step 1), tree and find (Step 2), find, wc, and head (Step 3), and find, wc, sort, and head (Step 4).

file-analysis