The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted external data in the form of interview transcripts and uploaded documents, creating a potential vector for indirect prompt injection.\n
Ingestion points: stages/intake.py contains log_interview and ingest_document methods that capture external text and file paths for processing.\n
Boundary markers: The ingestion logic does not currently utilize explicit delimiters or instructions to ignore nested prompts within the source data.\n
Capability inventory: The skill possesses capabilities to write to the local filesystem for persisting engagement journals, rendering markdown deliverables, and populating extraction queues and knowledge graph stubs (core/engagement.py, core/orchestrator.py, core/extraction/pipeline.py).\n
Sanitization: While a comprehensive anonymization pipeline (core/anonymize.py) strips personal and tenant markers, it does not perform sanitization of content for adversarial instructions.\n- [REMOTE_CODE_EXECUTION]: The IP extraction pipeline performs dynamic module loading by modifying sys.path to include a local path derived from the user's home directory. Additionally, the documentation suggests running remote scripts via npx during installation.\n
Evidence: In core/extraction/pipeline.py, the _bookkeeping_module function appends ~/broomva/skills/bookkeeping/scripts to sys.path and attempts to import the bookkeeping module.\n
Evidence: README.md provides an installation command using npx skills add broomva/skills.\n- [EXTERNAL_DOWNLOADS]: The skill's documentation and build configurations reference external repositories and resources owned by the author context.\n
Evidence: README.md and SKILL.md reference github.com/broomva/skills for cloning and methodology reference.

phronesis