The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest data from external URLs and use an LLM to summarize and synthesize this information into a persistent wiki. This creates a surface for indirect prompt injection attacks where malicious instructions embedded in a processed web page could influence the agent's behavior during the synthesis phase.
Ingestion points: Web content is fetched and saved to the raw/sources/ directory using scripts/ingest-url.sh via the scrapling tool.
Boundary markers: The workflow lacks explicit boundary markers or delimiters when presenting ingested raw content to the agent for synthesis, which increases the risk that embedded instructions are obeyed.
Capability inventory: The skill has access to Bash, Read, Write, Edit, Glob, Grep, and WebFetch, allowing the agent to modify project files and directory structures.
Sanitization: There is no evidence of content sanitization or filtering of the ingested web data before it is presented to the agent for synthesis.
[DYNAMIC_EXECUTION]: The skill utilizes small Python scripts embedded within Bash scripts (scripts/ingest-url.sh and scripts/new-query-note.sh) to perform file modifications, such as inserting lines at specific markdown markers. These snippets are executed at runtime using the python3 - command.
Evidence: The insert_marker_line function in scripts/ingest-url.sh and scripts/new-query-note.sh passes hardcoded Python source code via a heredoc to the Python interpreter to modify index.md and log.md files.

llm-wiki