The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted content from external GitHub repositories via the fetch-github command. This content is later processed by Haiku subagents in the generate-descriptions workflow to create metadata descriptions.
Ingestion points: The fetch-github command downloads README and documentation files from arbitrary GitHub repositories into the research/ directory.
Boundary markers: The skill provides specific formatting rules for the generated output, but does not specify delimiters or sanitization for the external input data when passed to subagents.
Capability inventory: The skill has the ability to write files (add, set-description), execute shell commands (uv run), and run remote tools (uvx).
Sanitization: No explicit sanitization or escaping of the fetched README content is mentioned before it is processed by subagents.
[COMMAND_EXECUTION]: The skill makes extensive use of shell commands via uv run to execute the local knowledge-explorer.py script and the gh CLI to interact with GitHub. It also invokes the system's $EDITOR for manual entry updates.
[EXTERNAL_DOWNLOADS]: The fetch-github command fetches repository metadata, release tags, and README content from GitHub's official API. While GitHub is a well-known service, this involves downloading content from third-party repositories.
[REMOTE_CODE_EXECUTION]: The generate-descriptions workflow instructs subagents to run uvx skilllint@latest. Using uvx to execute a remote package introduces a potential supply chain risk, although uvx is a standard tool for running packages from registries.
[DYNAMIC_EXECUTION]: The skill relies on uv run to execute its primary management script and utilizes uvx for dynamic tool execution. The workflow involves generating description strings that are then used in shell commands.

knowledge-explorer