The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill provides instructions to download and install the Bun runtime and the uv package manager from their official domains (bun.sh and astral.sh). It also directs the agent to clone a repository from an unknown and untrusted GitHub account (github.com/sdyckjq-lab/llm-wiki-skill).
[COMMAND_EXECUTION]: The installation process requires executing a local shell script (install.sh) from an untrusted repository, which could lead to arbitrary code execution. The skill also uses package runners like npx and uvx to execute remote extraction tools (baoyu-url-to-markdown, youtube-transcript).
[METADATA_POISONING]: There is a conflict between the stated author (ara.so) in the documentation and the actual vendor context (reason-machines). This discrepancy in metadata can be misleading regarding the skill's origin and security.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from diverse sources, which is a known vulnerability surface.
Ingestion points: Data is extracted from external web articles, tweets, and YouTube transcripts into the raw/ directory as markdown files (specified in SKILL.md).
Boundary markers: The instructions do not specify the use of delimiters or safety prompts to isolate processed external content from the agent's system instructions.
Capability inventory: The agent possesses the capability to execute shell commands (bash, npx, uvx) and perform file system writes.
Sanitization: There are no documented sanitization or filtering steps applied to the extracted content before it is synthesized into wiki pages.

llm-wiki-skill