The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The Python scripts utilize dynamic loading to execute code from filesystem paths computed at runtime:
scripts/architecture_graph.py, scripts/projection_engine.py, and scripts/rule_projection.py use importlib.util to load and execute modules (such as tickets.py) from relative paths within the project structure.
[COMMAND_EXECUTION]: The skill manifest and documentation describe operations that invoke the psql command-line utility:
Entrypoints are defined to run psql for initializing schemas and applying migration files (e.g., init.sql, migrate-v7-behavior-graph.sql).
Documentation recommends executing these commands with the high-privilege postgres superuser account.
[PROMPT_INJECTION]: The skill processes repository content into documentation, which constitutes an indirect prompt injection surface:
Ingestion points: Data is read from files within skills/, note/, review/, spec/, and docs/ directories.
Boundary markers: The projection engine uses  and  markers in its generated markdown files.
Capability inventory: The skill possesses capabilities for shell command execution (psql) and filesystem write operations.
Sanitization: There is no evidence of content sanitization or validation for ingested file data prior to processing or projection.

skill-system-postgres