The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses shell commands to manage a research vault within the ./research/ directory. These include git operations (status, diff, add, commit, push) and uv run for internal Python scripts. These commands are integral to the skill's primary function of maintaining a structured research corpus.
[DYNAMIC_EXECUTION]: The validate_research.py script utilizes subprocess.run() to execute git rev-parse --show-toplevel. This is a benign operation used to programmatically discover the repository root during the validation process.
[EXTERNAL_DOWNLOADS]: The skill includes instructions (repo-access-procedure.md) for agents to use git clone to fetch resources from GitHub for research. As GitHub is a well-known service and the downloads are for research documentation, this is a standard and safe practice.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from external URLs, which presents a surface for indirect prompt injection. However, it implements extensive mitigations, including a Validation Gate that uses automated scripts (validate_research.py, fix_research_formatting.py) and an Entry Review Rubric that employs a reviewer-worker agent loop to audit and correct content against quality standards before it is finalized.

research-curator