The Agent Skills Directory

[DATA_EXPOSURE]: The skill reads configuration from ~/.claude/settings.json and project-specific .claude/settings.json files. This access is restricted to retrieving internal configuration keys (rememberFolder, primaryProject) to determine where notes should be stored. There is no evidence of these settings or any discovered secrets being exfiltrated or used for unauthorized purposes.
[INDIRECT_PROMPT_INJECTION]: The skill accepts untrusted data through the $ARGUMENTS parameter and conversational context to generate notes.
Ingestion points: The $ARGUMENTS variable and conversation history in SKILL.md.
Boundary markers: None; the skill does not wrap the content in delimiters before passing it to the storage tool.
Capability inventory: The skill utilizes the write_note tool to perform persistent file-writing operations.
Sanitization: The instructions mandate that the agent capture content "verbatim" and refrain from adding its own observations or structure, which acts as a primary control to prevent the agent from being influenced by potentially malicious instructions within the captured text.

bm-remember