The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill's setup documentation instructs users to install the tool by piping a remote script from GitHub directly into a bash shell (bash <(curl -sL https://raw.githubusercontent.com/rarce/git-wiki/main/install.sh)). This is a high-risk pattern that permits arbitrary code execution from an external source.- [EXTERNAL_DOWNLOADS]: The scripts/setup.sh script automatically performs a global installation of the @tobilu/qmd package via npm. This introduces a third-party dependency from an external registry that is not part of the core skill or a verified vendor.- [PROMPT_INJECTION]: The skill's primary function is to 'ingest' untrusted external content (articles, papers, URLs) into the agent's context for synthesis, creating an attack surface for indirect prompt injection. * Ingestion points: External content is fetched and saved to the sources/ directory via the ingest operation in SKILL.md. * Boundary markers: No delimiters or instructions are provided to the agent to treat ingested content as untrusted data. * Capability inventory: The skill has extensive shell access, including git, gh, qmd, and npm. * Sanitization: No validation or sanitization of the ingested markdown content is performed before the agent processes it.- [COMMAND_EXECUTION]: The skill relies on complex shell interactions for its operations. The use of agent-generated topics in search queries (qmd query "<topic>") and the automated installation of packages create opportunities for command injection if the agent is manipulated via malicious external input.

git-wiki