self-improving-agent
Pass
Audited by Gen Agent Trust Hub on Apr 3, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill implements an autonomous self-improvement loop that creates a surface for Indirect Prompt Injection.
- Ingestion points: Feedback data is ingested from external logs (JSON/JSONL) and memory files via
scripts/feedback_analyzer.pyandscripts/rule_manager.py, as well as the manual review workflow described inSKILL.md. - Boundary markers: No boundary markers, delimiters, or instructions to ignore embedded prompts are defined for the processed feedback data.
- Capability inventory: The skill directs the agent to update
CLAUDE.mdand.claude/rules/, which are high-authority configuration files that determine the agent's behavior in future sessions. - Sanitization: The skill lacks sanitization, validation, or escaping of textual content in feedback logs before it is promoted to persistent behavioral rules.
Audit Metadata