The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses uv run --frozen to execute various local CLI tools including treg, agent-reach, and opencli for research, catalog lookups, and testing.
[INDIRECT_PROMPT_INJECTION]: The skill fetches data from public social media platforms (Reddit and X) using search queries which could contain malicious instructions.
Ingestion points: Data enters the environment via opencli reddit search and opencli twitter search commands executed in SKILL.md.
Boundary markers: The instructions explicitly direct the agent to 'delegate the reading to a subagent so the raw dump never enters the main context', providing a degree of isolation.
Capability inventory: The skill has capabilities to execute shell commands via uv run, write to local project files (src/treg/agent_pages.py), and perform network operations through specialized CLI tools.
Sanitization: The skill instructs the agent to 'filter vendor astroturf' and exclude marketing-heavy or repetitive vendor posts from the research output.

treg-page