focused-fix

Pass

Audited by Gen Agent Trust Hub on Nov 8, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied bug descriptions to identify relevant code areas. Ingestion points: The --bug parameter in scripts/change_scope_analyzer.py accepts free-text input. Boundary markers: The workflow in SKILL.md includes mandatory verification steps such as reviewing diffs and running regression tests. Capability inventory: The scripts/change_scope_analyzer.py script performs file-system reads across the codebase to score relevance but has no write or network capabilities. Sanitization: The analyzer uses a restrictive regex-based keyword extraction method (re.findall(r'\\b[a-zA-Z_]{2,}\\b')), which discards non-alphanumeric characters and command structures, effectively neutralizing most prompt injection payloads.
  • [COMMAND_EXECUTION]: The skill utilizes a local Python utility for scope analysis. The scripts/change_scope_analyzer.py script is implemented using only standard Python libraries. Analysis of the script confirms it does not perform network operations, execute arbitrary shell commands, or access sensitive system paths (it explicitly skips directories like .git and node_modules).
  • [SAFE]: The skill promotes secure development practices by enforcing a 'focused-fix' methodology that prioritizes minimal changes and mandatory regression tests. The provided documentation and anti-patterns define clear boundaries for the agent's behavior, preventing unauthorized scope creep.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 8, 2026, 03:14 PM