oss-forensics-orchestration

Pass

Audited by Gen Agent Trust Hub on Jun 7, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes a local initialization script using Bash to prepare the environment and check for credentials. Evidence: Phase 0 logic executes source .venv/bin/activate && python .claude/skills/oss-forensics/github-evidence-kit/scripts/init_investigation.py.
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it ingests untrusted data from user prompts (repository references, actor names, and vendor report URLs) and agent-generated responses without sanitization or boundary markers. 1. Ingestion points: Phase 1 (parsing user prompt for repositories and actors) and Phase 3 (parsing agent-generated evidence requests). 2. Boundary markers: No explicit delimiters or instructions are used to separate untrusted content from instructions. 3. Capability inventory: Bash script execution, file read/write, and dynamic sub-agent orchestration. 4. Sanitization: No escaping or validation is performed on ingested strings before interpolation into sub-agent prompts.
  • [DYNAMIC_EXECUTION]: The skill dynamically determines which sub-agent to invoke and what query to execute based on data parsed from files generated at runtime. Evidence: Phase 3 logic extracts an agent_name from a temporary file and uses it to spawn a new task via Task: {agent_name}.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 7, 2026, 01:29 PM