The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill recommends that the user install and the agent utilize an external search tool (qmd) from a personal GitHub repository (github.com/tobi/qmd) to handle large-scale wiki search operations.
[COMMAND_EXECUTION]: The instructions explicitly direct the agent to "shell out" to the qmd binary, executing search commands against the local filesystem based on user-provided queries.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to ingest external documents, URLs, and text pastes into a compounding knowledge base, which creates a significant surface for indirect prompt injection attacks.
Ingestion points: External source documents placed in the raw/ directory, content fetched from arbitrary URLs provided by the user, and direct text pastes in chat.
Boundary markers: Absent; the skill uses markdown headers for structure but does not implement delimiters or system-level instructions to ignore potential commands embedded within the ingested source text.
Capability inventory: The skill possesses Write and Edit permissions to modify the wiki filesystem and utilizes shell execution for search operations, creating a path for malicious data to influence agent behavior.
Sanitization: Absent; the skill does not include any steps to filter, escape, or sanitize content from ingested sources before incorporating them into the persistent wiki pages.
[DATA_EXFILTRATION]: The Ingest mode workflow instructs the agent to fetch content from user-provided URLs. This involves performing network requests to arbitrary, non-whitelisted domains which could be used for simple data exfiltration if the URL includes sensitive parameters.

llm-wiki