The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the dynamic context injection syntax (!command) to run a local utility script, pr-provenance-stamp.js. The command uses environment variables (CLAUDE_SKILL_DIR, CLAUDE_PLUGIN_ROOT) to locate the script, ensuring it only executes code from the plugin's own directories.
[PROMPT_INJECTION]: The skill processes session ledger data via the script's output to generate a summary. While this constitutes an indirect input surface, the task is restricted to summarizing metadata (tests and line counts) and the skill explicitly disables further model tool invocation via frontmatter configuration. Its purpose is to report on automated activities rather than execute instructions from untrusted data.

receipt