ai-scaling-laws
Pass
Audited by Gen Agent Trust Hub on Oct 6, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill maintains a persistent state via 'learnings.md' and 'learnings.consolidated.md', which are read at load time to inform agent behavior. This creates a potential vector where data from previous interactions could influence the agent's logic in future sessions.\n
- Ingestion points: The agent is instructed to read 'learnings.md' and 'learnings.consolidated.md' (SKILL.md), and to verify facts against 'data/sources.json'.\n
- Boundary markers: There are no explicit delimiters or instructions to ignore potential commands embedded within the learning logs or source metadata.\n
- Capability inventory: The skill executes a local Python script ('append_learning.py') and is instructed to verify external facts, potentially using network-enabled tools.\n
- Sanitization: No sanitization or validation logic is defined for the content of the learning files.\n- [COMMAND_EXECUTION]: The skill instructions specify a feedback loop that executes a local script: 'python3 agents-skills-feedback-loop/scripts/append_learning.py'.\n- [EXTERNAL_DOWNLOADS]: The skill references numerous external research papers and technical blogs from well-known organizations including ArXiv, Meta, OpenAI, DeepMind, and Epoch AI. These references are used for fact-checking and provide the mathematical basis for the skill.
Audit Metadata