Skill Install Safety Gate

Automated security and compatibility firewall for installing AI agent skills and Codex/OpenClaw packages.

Install
cmdop skills install agensi-skill-install-safety-gate

Skill Install Safety Gate inspects third-party skills and toolsets for Codex and OpenClaw-style architectures before they are written to the file system. It runs static analysis to verify package shape, ZIP integrity, and the presence of mandatory SKILL.md metadata. The capability checks the local host for required binaries listed under anyBins, along with required environment variables and configuration dependencies. It scans for suspicious patterns including ZIP path traversals, obfuscated payloads, prompt-injection risks, and dangerous dynamic code execution. It also checks whether a skill slug already exists in the target directory to prevent accidental overwrites. The skill operates through a programmatic pass-or-fail mechanism and generates structured JSON and CSV reports for agent or pipeline consumption. It outputs an install-safety-report.md file for human review and an output.json file for machine readability. If invoked with the —install-on-pass flag, it commits files to the target directory only when the gate returns a clean allow decision with zero blockers or review items. It never executes untrusted code during the audit.

Use cases

  • Use it to validate ZIP integrity and mandatory SKILL.md metadata before admitting a new skill into an agent runtime.
  • Use it to scan skill packages for ZIP path traversals, obfuscated payloads, prompt-injection risks, and dangerous dynamic execution patterns.
  • Use it to verify that the local host provides required binaries, environment variables, and config dependencies declared by a skill package.
  • Use it to prevent accidental overwrites by checking for existing skill slugs in the target directory before installation.
  • Use it to generate structured JSON and CSV reports that feed an automated CI/CD pipeline or agent decision loop.
  • Use it to conditionally install a skill only when the gate issues a clean allow decision with zero blockers.

When to use it

  • You need a programmatic pass-or-fail gate for admitting third-party skills into a Codex or OpenClaw agent runtime.
  • You want static analysis of skill packages without executing untrusted code during the audit.
  • Your workflow requires machine-readable JSON or CSV reports to drive downstream automation.
  • You need to verify host compatibility against declared binary, environment, and config dependencies.
  • You want human-readable markdown reports alongside machine-readable output for mixed review workflows.

When not to use it

  • You need runtime behavior analysis or dynamic execution of the skill code rather than static inspection.
  • You are installing non-skill artifacts that do not follow Codex or OpenClaw package conventions.
  • You require network-level sandboxing or OS-level container isolation, which this skill does not provide.
  • Your package format is not based on ZIP archives or does not include SKILL.md metadata.
  • You need to audit MCP servers or other server-style capabilities rather than skill packages.