The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill creates and executes scripts in Python (dag_py.py), R (dag_r.R), and Mermaid (dag.mmd) based on user-provided theory and labels. These are rendered into visual formats using system interpreters as described in phases/phase3-mermaid.md, phases/phase4-r.md, and phases/phase5-python.md.
[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands to produce diagram files. This includes using mmdc (Mermaid CLI), Rscript for R processing, and uv for Python execution with inline dependencies.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze external research papers (PDFs or full text) to extract causal claims in phases/phase0-theory.md. This represents a potential vector where instructions hidden in the analyzed documents could influence the agent's behavior.
Ingestion points: Research papers and theory notes provided in Phase 0; node labels provided in Phase 2.
Boundary markers: None identified; the instructions do not include specific delimiters or warnings to ignore instructions within the source documents.
Capability inventory: File system write access (creating scripts) and subprocess execution (uv, Rscript, mmdc) across multiple rendering phases.
Sanitization: There are no instructions to sanitize or escape node labels or descriptions before they are inserted into the Python and R script templates, which could lead to code injection if labels contain malicious control characters.

dag-development