self-improving-agent-builder
Pass
Audited by Gen Agent Trust Hub on Nov 2, 2026
Risk Level: SAFEDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [DYNAMIC_EXECUTION]: The skill enables an automated 'IMPROVE' phase where the agent generates and applies 'Code logic fixes' to its own source code, which is then executed during subsequent evaluation iterations.
- [INDIRECT_PROMPT_INJECTION]: The skill derives code and prompt improvements based on evaluation results and failure patterns, which may be influenced by external data processed during benchmarking.
- Ingestion points: Evaluation scores and failure logs processed in the ANALYZE and RESEARCH phases.
- Boundary markers: No explicit delimiters are defined to isolate test data from the analysis logic.
- Capability inventory: File system modifications (commit/revert) and command execution via the amplihack CLI.
- Sanitization: Employs a 'RESEARCH' phase that requires hypothesis testing and evidence gathering before applying any changes.
- [COMMAND_EXECUTION]: The skill instructs the agent to run various command-line tools such as 'python -m amplihack.eval.progressive_test_suite' and 'amplihack eval run' to manage the iteration cycles.
- [EXTERNAL_DOWNLOADS]: The skill references the vendor's repository at 'https://github.com/rysweet/amplihack' as the source for the core evaluation framework and agent SDK implementations.
Audit Metadata