The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches content from Anthropic's official documentation at https://code.claude.com/docs/en/goal to extract current character limits and formatting rules. This is a well-known service and the fetch is used for configuration purposes.
[COMMAND_EXECUTION]: The skill executes a local shell script goal-condition-length.sh located within the plugin root. It uses a robust method for passing user-controlled content by writing it to a temporary file (--file <path-to-draft>) instead of direct shell interpolation, which prevents common shell injection attacks from the goal text itself.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface because it parses external, untrusted content (the documentation website) to determine the value for the --limit argument passed to the shell script. If the documentation source were compromised or malformed, the extracted value could potentially be used to inject shell metacharacters into the command line.
Ingestion points: https://code.claude.com/docs/en/goal (Step 1).
Boundary markers: None; the skill relies on the AI to accurately extract a number from the page text.
Capability inventory: Execution of a local Bash script via bash "${CLAUDE_PLUGIN_ROOT}/scripts/goal-condition-length.sh" (Step 3).
Sanitization: The skill explicitly recommends using the --file flag to sanitize the drafted condition content, but it lacks specific validation or sanitization instructions for the --limit value extracted from the web.

draft-goal-condition