taku-reflect
Pass
Audited by Gen Agent Trust Hub on May 7, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses the Bash tool to execute standard git commands (git log, git config, git shortlog) for computing metrics and its own Python management script (scripts/learnings.py) to handle JSONL data operations.
- [PROMPT_INJECTION]: Implements an indirect prompt injection surface through its 'Auto-Recall' and 'Learn' mechanisms, which gather data from the project environment and user dialogue to influence future agent context. \n
- Ingestion points: Git commit history and author metadata gathered during retrospectives, and user-provided feedback recorded in .taku/learnings/. \n
- Boundary markers: Includes the TAKU_LEARNINGS_PROTOCOL block which explicitly instructs the agent to treat retrieved learnings as 'context, not hard rules'. \n
- Capability inventory: The skill has access to the Edit and Write tools, allowing it to modify project instruction files (AGENTS.md, CLAUDE.md), and Bash for executing local scripts. \n
- Sanitization: Content is stored in JSONL format using standard serializers, but no natural language sanitization is performed on the ingested text before it is presented to the agent in subsequent sessions.
Audit Metadata