The Agent Skills Directory

[PROMPT_INJECTION]: The skill's metadata contains deceptive claims regarding its origin and safety. It identifies as an 'official' source from 'DAIR.AI' and explicitly labels itself as 'safe' in its frontmatter, which contradicts the provided author context. This misrepresentation could lead users to grant the skill excessive trust based on false reputation.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection when processing untrusted data.
Ingestion points: The agent is instructed to read contents from a raw/ directory and external URLs defined in sources.md as part of its synthesis workflow.
Boundary markers: There are no protective delimiters (e.g., triple quotes or XML tags) or specific instructions in the templates/prompts/ files to isolate source content from agent instructions.
Capability inventory: The skill allows for local file system manipulation and script execution via the bundled init_wiki.sh script.
Sanitization: No validation or filtering is applied to the ingested source data before it is processed by the agent.
[COMMAND_EXECUTION]: The skill uses a bash script (scripts/init_wiki.sh) to perform file system operations. While it includes basic validation for directory names using a regex, it relies on sed for template interpolation of user-provided strings like the wiki title, which can lead to unexpected behavior if special characters are not handled correctly.

wiki-builder