The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and summarize untrusted multi-source documents including Slack threads, customer emails, and meeting transcripts, which introduces an indirect prompt injection vulnerability surface.
Ingestion points: External inputs are loaded and processed during "The Operation" phase described in SKILL.md.
Boundary markers: There are no explicit text boundaries or instruction isolation markers specified to contain untrusted data.
Capability inventory: The skill contains capabilities to write and update files (SPEC.md, .memlog.md, companions) and invoke local automation scripts via uv run.
Sanitization: While the skill implements filtering rules to maintain lean prose and content relevance, it does not explicitly sanitize or escape data to mitigate hidden text directives.
[COMMAND_EXECUTION]: The skill performs command execution sequences using uv run to launch local helper scripts (resolve_customization.py, resolve_config.py, and memlog.py) located under the project's internal directory (_bmad/scripts/). It also includes support for optional workflow hooks (activation_steps_prepend, activation_steps_append, and on_complete) defined within configuration files to automate setup and completion tasks.

bmad-spec