The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches content from the official documentation repository camunda/camunda-docs using gh repo clone and the GitHub API. These operations are directed at the vendor's own infrastructure and are necessary for documentation management.
[COMMAND_EXECUTION]: Employs standard system and developer tools (git, gh, mktemp, rm) to manage a temporary workspace, analyze changes, and synchronize documentation. The use of mktemp for a temporary directory and subsequent cleanup is a standard safe practice.
[OBFUSCATION]: The skill uses base64 -d to decode responses from the GitHub API. This is the standard method for handling file content returned by the GitHub API's repository contents endpoint. The target file is a documentation style guide within the vendor's repository.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: Untrusted data enters the agent context via git diff, git log, and gh pr view (PR title and body) in SKILL.md.
Boundary markers: No explicit delimiters or instructions are used to separate ingested data from agent instructions in the prompts.
Capability inventory: The skill can perform file writes, git commit, git push, and manage PRs via gh pr create and gh pr edit as seen in SKILL.md.
Sanitization: There is no evidence of sanitization or filtering of the ingested external content before it is processed by the agent.
Risk Assessment: While an injection surface exists where code comments or PR descriptions could influence the documentation output, the operations are limited to the specific task of documentation generation within the authorized vendor repository.

write-camunda-docs