heal-skill

Pass

Audited by Gen Agent Trust Hub on Mar 5, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill implements a workflow to modify instruction files (SKILL.md) based on observations from conversation history. This introduces a risk of Indirect Prompt Injection, where an attacker could provide malicious "corrections" in the chat to trick the agent into inserting unauthorized instructions into its own skills.
  • Ingestion points: Interprets conversation context and existing SKILL.md files to derive code and instruction changes.
  • Boundary markers: The skill does not define specific delimiters to separate untrusted data from the instructions used to generate the new SKILL.md content.
  • Capability inventory: Uses the 'Edit' tool to overwrite instruction files and 'Bash' to commit changes to a repository.
  • Sanitization: Relies on manual human review via the 'AskUserQuestion' tool before any modifications are applied.
  • [COMMAND_EXECUTION]: Executes shell commands via Bash to list files (ls) and manage version control (git). These operations are used for discovering skills and committing the generated fixes to the filesystem.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 5, 2026, 08:18 AM