simplify-and-harden

Pass

Audited by Gen Agent Trust Hub on Oct 27, 2026

Risk Level: SAFEPERSISTENCEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [PERSISTENCE]: The 'Learning Loop' feature automatically promotes recurring quality and security patterns into the agent's core context files (such as CLAUDE.md, AGENTS.md, or .github/copilot-instructions.md). This mechanism allows for the persistent modification of the agent's system-level instructions based on data processed during previous tasks.
  • [INDIRECT_PROMPT_INJECTION]: The skill analyzes code diffs and modified files, which are potentially untrusted inputs. Maliciously crafted code or comments could attempt to influence the agent's reasoning during the Simplify and Harden passes, especially since the skill lacks explicit boundary markers to separate the content being reviewed from the agent's instructions.
  • Ingestion points: Analyzes code diffs and modified source files across various languages (e.g., .ts, .py, .go) as defined in SKILL.md.
  • Boundary markers: None identified. The skill instructs the agent to read the code directly without specifying delimiters or 'ignore embedded instructions' warnings.
  • Capability inventory: The skill can modify source files, write to a learning log (.learnings/LEARNINGS.md), update agent context files, and execute shell commands for verification.
  • Sanitization: No explicit sanitization or validation of the code-under-review is mentioned before it is processed by the agent's logic.
  • [COMMAND_EXECUTION]: The re-verification pass (Pass 4) requires the agent to run shell commands (e.g., npm test) on the modified codebase. While standard for development, this could be leveraged to execute malicious code if an attacker has influenced the test scripts or the files being verified.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 27, 2026, 09:23 PM