skill-builder-by-rdd

Pass

Audited by Gen Agent Trust Hub on Sep 20, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSCREDENTIALS_UNSAFE
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process user-provided materials such as laws, tables, and case examples during the discovery phase. This creates a surface for indirect prompt injection if those documents contain malicious instructions. The skill proactively addresses this in its security documentation (references/SECURITY.md), instructing the agent to treat all uploaded content strictly as data and evidence, and explicitly forbidding the execution of commands found within attachments.
  • Ingestion points: Discovery and Collection phases in SKILL.md where external materials are cataloged.
  • Boundary markers: Explicit instructions in references/SECURITY.md and SKILL.md to ignore 'ignore previous rules' type commands in data.
  • Capability inventory: Local file reading for linting purposes and execution of validation scripts using the standard Python library.
  • Sanitization: The scripts/lint_bundle.py script includes checks for null bytes and specific secret patterns.
  • [COMMAND_EXECUTION]: The skill bundle includes Python scripts (lint_bundle.py, scope_gate.py) used for validating the structure and consistency of generated skills. The unit tests in tests/test_controls.py utilize subprocess.run to invoke these local scripts during the testing phase. This is an expected development practice for tool verification and does not involve executing untrusted network-sourced commands.
  • [EXTERNAL_DOWNLOADS]: The skill's documentation (references/RUNTIME_COMPATIBILITY.md) contains links to official specification pages and help articles from well-known services such as agentskills.io, OpenAI, and Anthropic. These references are used for providing platform-specific context and are considered safe.
  • [CREDENTIALS_UNSAFE]: The skill contains a security linter (scripts/lint_bundle.py) that implements regex-based detection for sensitive information, such as OpenAI API keys, AWS access keys, and private RSA keys. This feature is intended to prevent the accidental leakage of credentials during the skill development process.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 20, 2026, 07:20 PM