learn
Pass
Audited by Gen Agent Trust Hub on May 11, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: Executes various helper binaries located in
~/.claude/skills/gstack/bin/for session tracking, configuration management, and searching project learnings. - [COMMAND_EXECUTION]: Uses
bun -eto run an inline JavaScript snippet that parses and aggregates data fromlearnings.jsonl. - [EXTERNAL_DOWNLOADS]: Performs update checks and sends telemetry logs to remote services via internal binaries.
- [EXTERNAL_DOWNLOADS]: Includes a reference to a documentation URL (
https://garryslist.org/posts/boil-the-ocean) associated with the author's methodology. - [PROMPT_INJECTION]: The skill ingests and processes data from
learnings.jsonlduring search and prune operations. This creates a surface for indirect prompt injection, as external data is incorporated into the agent's context without explicit boundary markers or sanitization of potential instructions within the JSONL records.
Audit Metadata