The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user input (script purpose and chosen name) and interpolates it into file headers and system paths used for tool execution. This creates a surface where malicious input could influence the shell commands run during the validation step.
Ingestion points: User-provided text for script purpose and filename in SKILL.md Step 1.
Boundary markers: Absent for the interpolation of user-supplied paths into shell command arguments.
Capability inventory: The skill performs file system writes, modifies permissions via chmod +x, and executes system utilities like shellcheck and bash -n in SKILL.md Step 5.
Sanitization: The instructions recommend snake-casing for the script name but do not explicitly specify shell-escaping for the path before use in command-line arguments.
[DYNAMIC_EXECUTION]: The skill generates shell scripts from a local template (assets/template.sh) and user-provided configuration, then prepares them for execution.
Evidence: SKILL.md Step 5 describes writing the assembled script to a user-chosen path and setting the executable bit.
[COMMAND_EXECUTION]: The skill invokes local system utilities to retrieve environment data and validate the generated script's syntax.
Evidence: The skill executes git config user.name to identify the author (Step 4) and runs shellcheck and bash -n (Step 5) to verify the output.
[PRIVILEGE_ESCALATION]: The skill template includes functions for privilege escalation and the skill itself modifies file permissions to make scripts executable.
Evidence: assets/template.sh contains check_superuser and run_as_root functions utilizing sudo. SKILL.md Step 5 uses chmod +x on the generated script path.

bash-scaffold