invisible-threads
Pass
Audited by Gen Agent Trust Hub on Nov 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it ingests untrusted text data and processes it using an LLM without sufficient isolation.
- Ingestion points:
scripts/chunk_corpus.pyreads markdown files from a source directory into a database;scripts/extract_insights.pythen reads these segments. - Boundary markers: Absent. The
EXTRACTION_PROMPTinscripts/extract_insights.pyinterpolates{chunk_text}directly at the end of the prompt without delimiters or instructions to ignore embedded instructions. - Capability inventory:
scripts/extract_insights.pymakes network calls to Gemini and Claude APIs;scripts/find_threads.pyandscripts/extract_insights.pyexecute shell commands to install packages. - Sanitization: Absent. No filtering or escaping is applied to the corpus text before it is sent to the LLM.
- [COMMAND_EXECUTION]: The scripts use system calls to manage the execution environment.
- Evidence:
scripts/extract_insights.pyandscripts/find_threads.pyinvokesubprocess.check_call()to runpip installcommands. - [EXTERNAL_DOWNLOADS]: The skill dynamically fetches necessary libraries from public registries during runtime.
- Evidence: Fetches well-known packages including
google-generativeai,python-louvain, andsentence-transformersfrom PyPI.
Audit Metadata