The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local Python script memory-manager.py (located in the plugin's scripts directory) with various administrative actions including prepare-store, get-memory-status, and log-stored.
[DATA_EXFILTRATION]: The skill reads local files within ~/.claude-marketing/ to load active brand context, profiles, and compliance rules. It also checks for the presence of sensitive environment variables, such as PINECONE_API_KEY, to verify the configuration of external memory services.
[INDIRECT_PROMPT_INJECTION]: The skill exposes a surface for indirect prompt injection by ingesting and storing untrusted data from multiple sources, including external URLs and session content, which could be exploited when retrieved in future contexts.
Ingestion points: Processes 'Content to store' which can include user-provided text, session references, reports, or URLs to external research (SKILL.md).
Boundary markers: No specific boundary markers or 'ignore embedded instructions' warnings are mentioned for the content being saved to the vector database.
Capability inventory: The skill can perform local file writes to ~/.claude-marketing/, execute shell commands via memory-manager.py, and perform network operations through connected vector-DB MCPs like Pinecone or Qdrant.
Sanitization: Content undergoes normalization, SHA-256 hashing, and metadata structuring via the memory-manager.py script before storage.

save-knowledge