The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs users to install software packages from public registries where the publisher identity does not match the skill author.
Evidence: Python instructions require pip install honker.
Evidence: Node.js instructions require npm install @russellthehippo/honker-node. The scope @russellthehippo is inconsistent with the skill author 'brojonat', presenting a supply chain impersonation risk.
[DYNAMIC_EXECUTION]: The skill requires loading a compiled native shared library into the SQLite process.
Evidence: The instructions for raw SQLite include .load ./libhonker_ext. Native binary execution can bypass language-level security controls and execute arbitrary code on the host machine.
[INDIRECT_PROMPT_INJECTION]: The skill functions as a consumer for data stored in database tables, which acts as a vector for malicious instructions to influence agent behavior.
Ingestion points: Data is read from the payload column in _honker_live, _honker_streams, and _honker_notifications tables.
Boundary markers: The skill instructions do not specify the use of delimiters or 'ignore' instructions for the processed data.
Capability inventory: The skill allows for direct file system modification (via SQLite) and subprocess execution for background tasks.
Sanitization: No evidence of input validation or payload sanitization is present in the skill documentation.
[COMMAND_EXECUTION]: The documentation provides instructions for running background workers that execute arbitrary tasks based on database state.
Evidence: python -m honker worker myapp.tasks:db --queue=emails --concurrency=4

honker