The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructions direct the agent to download content from a third-party GitHub repository (github.com/Learning-Bayesian-Statistics/baygent-skills.git) which is not among the verified vendor resources.
[REMOTE_CODE_EXECUTION]: The setup process requires cloning the external repository and copying its contents into the agent's internal skills directory (~/.claude/skills/), effectively introducing new executable instructions into the agent's environment from an unverified source.
[COMMAND_EXECUTION]: The skill provides multiple shell commands for the agent to execute, including checking for dependencies (ls), cloning repositories (git clone), and installing Python packages (mamba install).
[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes external observational data for causal analysis, creating a surface for indirect prompt injection.
Ingestion points: User-provided data files or descriptions regarding marketing, medical, or policy outcomes.
Boundary markers: The skill implements mandatory user checkpoints for the Causal DAG and identification strategy, which provides some mitigation against accidental instruction following.
Capability inventory: The skill can execute shell commands for setup and use Python libraries (PyMC, CausalPy) to perform complex computations.
Sanitization: The instructions do not specify any explicit sanitization or filtering of the content within the ingested datasets.

causal-inference