T01 · Skill Instruction Hijacking
Error
- Location
scripts/bootstrap-skill-preflight.py:151- Finding
Persistent Agent Instruction Hijacking Through Repository Hooks and AGENTS.md
- Content
View full analysis
None: target = context.target_root / "AGENTS.md" if target.exists(): current = target.read_text(encoding="utf-8") if AGENTS_MARKER in current: context.log("SKIP AGENTS.md (already contains preflight section)") return context.log("APPEND AGENTS.md") if context.dry_run: return separator = "\n\n" if not current.endswith("\n\n") else "" target.write_text(current.rstrip() + separator + AGENTS_SECTION + "\n", encoding="utf-8") return context.log("WRITE AGENTS.md") if context.dry_run: return target.write_text(MINIMAL_AGENTS_HEADER + AGENTS_SECTION + "\n", encoding="utf-8") ``` ```bash #!/bin/bash set -e cat <<'EOF' Before processing the current task, run the Skill preflight process, use remote Skill discovery when local results are unsuitable, and report which Skills will be used. Record reusable experience and errors in the repository learning files. EOF ``` The final shell excerpt i ...[truncated 2827 chars]- Remediation
View remediation