backstage-scaffolder-template-review - SKILL.md Agent Skill

name: backstage-scaffolder-template-review description: Use this skill when reviewing Backstage Scaffolder software templates. Trigger when the user asks whether a template is safe for developer self-service, whether template RBAC gates are in place, whether input parameters are validated, whether a step action has excessive blast radius, or whether template outputs expose secrets. allowed-tools: Read Grep Glob metadata: author: "github: Raishin" version: "0.1.0" updated: "2026-05-05" category: delivery

Backstage Scaffolder Template Review

Purpose

Review Backstage Scaffolder Template kind resources for action blast-radius, input parameter injection risk, RBAC permission gate coverage, integration secret scope, catalog entity poisoning via catalog:register, and plaintext secret exposure in output: stanzas. Backstage Scaffolder gives developers a curated UI to trigger powerful backend actions — without RBAC gates and input validation, every authenticated developer effectively has write access to whatever the Scaffolder integration credentials can reach.

Lean operating rules

Prefer user-provided sanitized Template YAML as primary evidence; official Backstage docs are the authoritative fallback.
Treat any steps: action that provisions real cloud infrastructure (Terraform, Crossplane CRD apply, CloudFormation deploy, kubectl apply) with no RBAC permission gate as a CRITICAL finding.
Treat input parameters flowing unsanitized into publish:github.repoUrl, file-path actions, or shell-exec actions as a HIGH finding — path traversal and injection are realistic.
Treat publish:github with visibility: public as the default or without an allowedHosts constraint as a HIGH finding.
Treat output: stanzas exposing plaintext generated credentials, connection strings, or API keys in the Backstage UI as a HIGH finding.
Treat the absence of @backstage/plugin-permission-backend policies for infrastructure-provisioning templates as a HIGH finding — any authenticated Backstage user can trigger them.
Treat catalog:register accepting arbitrary user-supplied YAML without server-side entity schema validation as a MEDIUM finding — catalog poisoning overwrites ownership and lifecycle metadata.
Keep the answer scoped: report what was reviewed, the evidence level, and exactly which steps or fields triggered each finding.

References

Load these only when needed:

Workflow and output contract

Response minimum

Scoped target (Template metadata.name) and evidence level
Each steps: action type and its provisioning blast radius
Input parameter validation gaps (missing maxLength, pattern, enum)
RBAC permission gate verdict (present / absent / partial)
Integration secret scope assessment
output: stanza exposure assessment
Safe next actions and open questions