agent-analytics-autoresearch
Pass
Audited by Gen Agent Trust Hub on Oct 12, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONREMOTE_CODE_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted data from external analytics sources, creating an attack surface for indirect prompt injection.\n
- Ingestion points: Data snapshots retrieved from the Agent Analytics CLI and stored in the
data/directory, as well as parameters inbrief.md.\n - Boundary markers: While the skill follows a structured loop in
references/program.md, it does not utilize explicit delimiters to isolate external data from its instructions.\n - Capability inventory: The agent has permissions to execute shell scripts, run remote packages via
npx, and write to local files likeresults.tsv.\n - Sanitization: ANSI escape codes are stripped from terminal output using
perl, but the textual content of the analytics data is not further sanitized.\n- [COMMAND_EXECUTION]: Includes scripts for initializing the growth loop environment and automating data collection.\n scripts/init_autoresearch_run.shsets up the project structure.\nscripts/collect_agent_analytics_snapshot.shmanages the fetching of live metrics via the vendor's CLI.\n- [REMOTE_CODE_EXECUTION]: Usesnpxto download and execute the@agent-analytics/cli@0.5.31package from the npm registry. This is a vendor-owned resource used for its intended purpose of retrieving project analytics.
Audit Metadata