The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a Python script from the vendor's framework ({project-root}/_bmad/scripts/resolve_customization.py) during the initialization phase to resolve configuration overrides.
[COMMAND_EXECUTION]: The configuration structure in customize.toml includes activation_steps_prepend and activation_steps_append hooks. These are designed to execute arbitrary shell commands defined in the skill's defaults or project-specific overrides, allowing for dynamic command execution based on local configuration.
[COMMAND_EXECUTION]: The write-document.md and validate-doc.md instructions explicitly authorize the agent to spawn subprocesses for performing research, executing web searches, and reviewing documents.
[PROMPT_INJECTION]: The skill automatically loads untrusted data from the project environment via the persistent_facts mechanism in customize.toml, specifically targeting {project-root}/**/project-context.md.
Ingestion points: Project context files are loaded into the agent's foundational context in SKILL.md (Step 4).
Boundary markers: No explicit delimiters or instructions to ignore embedded commands are used during this ingestion.
Capability inventory: The skill has access to shell execution (Steps 1, 2, 7), subprocess spawning, and the ability to modify project files (update-standards.md).
Sanitization: No sanitization or validation of the ingested project context is performed before it is added to the agent's active memory.

gds-agent-tech-writer