The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill checks for the presence of two external dependencies: mattpocock-skills (specifically the grilling skill) and scribe (from the author's own repository healthier-vitamins/agentic-grimoire). While these require external installation using claude plugin install or npx, the skill explicitly prompts the user for consent before providing or executing these installation commands. These resources originate from either official curated registries or the vendor's own infrastructure.
[COMMAND_EXECUTION]: The skill performs several local command executions, including git rev-parse --is-inside-work-tree to determine repository context and reading/writing files within the current working directory. These operations are essential to its primary function of drafting documentation based on code and writing the resulting markdown files to the user's project.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data by reading files from the local repository to ground the ticket draft. This represents an indirect prompt injection surface where malicious instructions embedded in the codebase could theoretically influence the subagent's output. However, the risk is mitigated by the skill's narrow focus on generating documentation and the human-in-the-loop requirement for the final output.
[SAFE]: The use of subagents and model pinning (e.g., Claude Sonnet 5) is a standard platform capability used here to ensure output quality. All file access and network-adjacent operations (like package installations) are transparently disclosed in the instructions.

ticketsmith