The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements a persistent memory system that records conversation highlights into an append-only file (.memlog.md). Malicious instructions entered by a user or ingested from external data could be persisted and later re-activated when the agent distills these logs to 'shape the room' in future sessions.
Ingestion points: Found in references/party-memory.md (reading and distilling past memlogs) and references/create-party.md (distilling new personas from user-provided source data such as survey exports or interview notes).
Boundary markers: There are no defined delimiters or instructions to ignore embedded prompts within the ingested memory or source data content.
Capability inventory: The skill has the ability to execute shell commands via uv run, spawn subagents with independent context windows, and modify its own configuration using the bmad-customize tool.
Sanitization: The skill does not perform explicit sanitization, filtering, or validation of the text ingested from external sources or memory logs.
[COMMAND_EXECUTION]: The skill's architecture relies on executing internal Python scripts to resolve roster and configuration settings. The scripts/resolve_party.py script uses subprocess.run to call other scripts within the project infrastructure.
Evidence: SKILL.md defines several activation steps using uv run {project-root}/_bmad/scripts/.... The Python script scripts/resolve_party.py contains a _run_json function that executes external processes to fetch configuration data.

bmad-party-mode