sift-prime
Pass
Audited by Gen Agent Trust Hub on Sep 19, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes project files which could contain malicious instructions designed to influence the ticket generation process. * Ingestion points: The skill reads README.md, AGENTS.md, CLAUDE.md, and .ai/sift/MILESTONES.md (references/analysis.md). * Boundary markers: A 'hard scope fence' is defined in references/analysis.md to restrict the agent's focus. * Capability inventory: The skill writes to MILESTONES.md, creates directories, generates markdown ticket files, and executes local bash scripts (SKILL.md). * Sanitization: The orchestrator uses read-only sweep agents for analysis and a fixed template (references/drafting-agent-prompt.md) for writing to isolate untrusted data influence.
- [COMMAND_EXECUTION]: The skill relies on the execution of local shell scripts to manage its internal state and search for duplicates. * Evidence: SKILL.md and references/analysis.md include instructions to run scripts/existing-work.sh, scripts/reserve-ids.sh, and .ai/sift/scripts/sift.sh.
Audit Metadata