The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses local Python scripts scripts/wake.py and scripts/init-sanctum.py to manage its environment and load persistent memory. These are executed via the uv tool during skill activation and initialization.
[PERSISTENCE]: The skill implements a robust persistence mechanism by storing identity, long-term memory, and learned capabilities in a local directory structure within the project root (_bmad/memory/bmad-agent-creative-muse/). It includes a 'Pulse' mechanism for autonomous background maintenance and memory curation.
[DYNAMIC_EXECUTION]: The skill features 'Capability Authoring' where it can generate and save new capability prompts or scripts to its sanctum based on user interaction. While this allows the agent to evolve, it involves writing and subsequently reading/executing new logic at runtime.
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection as it is designed to ingest and curate data from session logs and project files into its long-term memory.
Ingestion points: Raw session logs (sessions/*.md) and files within the project root.
Boundary markers: The references/memory-guidance.md file provides philosophical guidance on what to remember, though the templates do not show strict structural delimiters for untrusted data.
Capability inventory: Local script execution (uv run), file writing to the sanctum directory, and dynamic prompt loading.
Sanitization: The skill relies on natural language instructions in memory-guidance.md rather than programmatic sanitization to filter data stored in memory.

bmad-agent-creative-muse