confirm

Pass

Audited by Gen Agent Trust Hub on Jul 30, 2026

Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes Git commands (git status, git diff, git branch, git log) to pre-compute project context and uses the !command syntax in the SKILL.md frontmatter. These are standard, benign operations for development tooling.\n- [DATA_EXFILTRATION]: While the skill aggregates information into evidence manifests, it explicitly defines a "redaction bar" that prevents raw command output, local paths, or credentials from being stored in persistent or committed files.\n- [PROMPT_INJECTION]: The skill ingests external data from conversation history and project files to verify intent and implementation.\n
  • Ingestion points: Conversation history (intent retrieval), git diffs (implementation inventory), and project files (e.g., CLAUDE.md).\n
  • Boundary markers: The skill uses a structured two-stage verification process but does not explicitly define delimiters for all interpolated text.\n
  • Capability inventory: Shell command execution (Git), file writing (manifests), and orchestration of other agent tools.\n
  • Sanitization: The skill implements a "redaction bar" for evidence manifests to ensure sensitive machine-local information is not exfiltrated or committed.\n- [SAFE]: No malicious patterns, obfuscation, or unauthorized data access were detected. The skill follows established best practices for secure tool orchestration and data handling.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 30, 2026, 07:55 PM
Security Audit — agent-trust-hub — confirm