jianying-narration

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 external ASR (Automatic Speech Recognition) transcripts and media metadata which are potentially attacker-controlled and could contain malicious instructions.\n
  • Ingestion points: Data enters the context via media.probe and media.asr_ledger outputs as referenced in SKILL.md and references/workflow.md.\n
  • Boundary markers: The skill implements 'protected speech boundaries' and a 'protect list' for specific entities (names, numbers, products) to maintain semantic integrity, as detailed in references/workflow.md.\n
  • Capability inventory: Capabilities include timeline.edit, media.probe, and media.asr_ledger for modifying video project files.\n
  • Sanitization: The workflow requires structure verification (project verify) and manual playback confirmation (playback) to validate that edits match the intended plan.\n- [COMMAND_EXECUTION]: The skill utilizes a local Rust-based CLI (jianying) to execute timeline operations and media analysis.\n
  • Evidence: Instructions in SKILL.md detail the use of jianying CLI for all processing steps.\n
  • Guardrails: The skill explicitly forbids overwriting source files, requiring all edits to be performed on isolated copies. It also mandates explicit user approval for any 'external_native_execution' and requires capability checks before command execution.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 19, 2026, 11:55 PM
Security Audit — agent-trust-hub — jianying-narration