The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local Python script {project-root}/_bmad/scripts/resolve_customization.py during both activation and completion phases to handle configuration logic and post-process hooks.
[COMMAND_EXECUTION]: The workflow performs dynamic execution of shell commands defined in the activation_steps_prepend and activation_steps_append fields. These commands are sourced from configuration files in the skill root and the project's _bmad/custom/ directory.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests untrusted data from the game codebase and project documentation to drive test generation.
Ingestion points: Reads all files matching the glob {project-root}/**/project-context.md and analyzes source code in src/ or engine-specific directories.
Boundary markers: No clear delimiters or instructions are used to separate ingested data from the system prompt.
Capability inventory: Includes file system writes ({test_dir}), command execution (activation steps), and script execution (resolve_customization.py).
Sanitization: There is no evidence of content sanitization or validation of the files loaded into the persistent facts context.

gds-test-automate