How bumper compares
bumper isn't trying to replace your scanner suite. It occupies a specific niche: the moment
of change — a terraform apply, a package install — and especially inside the agent
loop, where code is generated and run without a human in the path. This page is an honest
map of where it overlaps existing tools, where it's genuinely different, and where it
deliberately does less.
- Terraform plan safety
- Dependency safety
- The agent guardrail
- What bumper is not
- When to reach for bumper
Terraform plan safety
vs Checkov, Trivy/tfsec, Terrascan, KICS, Snyk IaC.
Those are excellent, mature config scanners with huge rule sets across many frameworks. They
overwhelmingly check the end state — "is this configuration misconfigured?" bumper's wedge
is different: it reads the plan diff — the create / update / delete / replace
transition — so it can catch the class of problem an end-state scanner structurally can't:
"this apply will destroy your database."
| bumper | Checkov / Trivy / Terrascan / Snyk IaC | |
|---|---|---|
| Reads the plan diff (create/delete/replace) | ✅ the core idea | ◑ mostly end-state config |
| Catches destruction (DB replace w/o snapshot, deletion-protection removed) | ✅ | ✗ (not a config-state property) |
| Enforces — binds a pass to the plan by sha256 and blocks the apply | ✅ verify + guard | ◑ warns / can fail CI; no apply-time bind |
| Agent tool-layer hook (blocks an unverified apply mid-session) | ✅ | ✗ |
| Deterministic, single static binary, fully offline | ✅ | ◑ varies |
| Raw rule count / frameworks | 112 enforced TF rules (+ ~2,600 advisory) | far more (thousands of rules; k8s, CFN, ARM, Helm, Dockerfile…) |
| Policy-as-code (Rego/OPA, custom graph policies) | ✗ (YAML+CEL rules) | ✅ (Checkov, others) |
The honest summary: for breadth — more clouds, more frameworks, thousands of rules, custom policy languages — the incumbents win, and bumper happily federates their catalogs for its advisory search. bumper wins on the transition (destruction/exposure as a change) and on enforcement in the agent loop. They're complementary: run Checkov/Trivy for broad config coverage, bumper as the apply-time safety gate.
Dependency safety
vs Dependabot, Snyk, Socket, osv-scanner, Trivy.
| bumper | Dependabot | Snyk | Socket | osv-scanner | |
|---|---|---|---|---|---|
| Known-vulnerable scan (OSV/CVE) | ✅ | ✅ | ✅ | ◑ | ✅ |
Known-malicious package block (OSV MAL-) | ✅ | ✗ | ◑ | ✅ | ◑ |
| Behavioral / zero-day malware detection | ✗ | ✗ | ◑ | ✅ (its specialty) | ✗ |
| Pre-execution block in the agent loop (deny the install before it runs) | ✅ | ✗ | ✗ | ◑ | ✗ |
| Auto-remediation version-bump PRs | ✗ | ✅ (its specialty) | ✅ | ◑ | ✗ |
| Lookup-not-upload (only coordinates leave; self-hostable) | ✅ | n/a (GitHub) | ✗ (SaaS) | ✗ (SaaS) | ✅ (local) |
| Same engine gates infra + deps | ✅ | ✗ | ◑ | ✗ | ✗ |
The honest summary: bumper's edge is the pre-execution malware block at the agent's tool
layer — it denies an install of a known-malicious package before the install runs, in the
loop where an AI agent would otherwise just run it. For known-malicious detection it relies
on OSV MAL- advisories; it does not do Socket-style behavioral analysis of novel
packages — Socket is more sophisticated there. For vulnerable-dependency remediation,
Dependabot's auto-PRs are the strength bumper doesn't have. Reach for bumper to stop the bad
install in the agent loop and in CI with one unified gate across infra and dependencies; keep
Dependabot/Snyk/Socket for what they each do best.
The agent guardrail
This is the part with the fewest direct comparables. bumper installs blocking pre-tool
hooks into Claude Code, Augment, and Gemini CLI so the agent cannot run an unverified terraform apply or install a known-malicious package — enforced at the tool layer, not as advice the
model can ignore. Most security tooling assumes a human runs the command and reads the report
later; bumper assumes an agent runs it, autonomously, and gates that. The hosted Advisor
MCP is the proactive half — the agent can ask "is this package safe?" / "what's the rule for
this resource?" before it writes the code. See agents.md.
What bumper is not
Being explicit, because a security tool that overclaims is worse than useless:
- Not a behavioral malware scanner. It blocks known-malicious packages (OSV
MAL-), not novel/zero-day payloads via static or behavioral analysis. - Not a dependency auto-updater. No version-bump PRs — pair it with Dependabot/Renovate.
- Not a full SCA platform. No license compliance, SBOM management, or reachability for dependencies (yet).
- Terraform only for IaC — no CloudFormation, Pulumi, Kubernetes manifests, or Helm today.
- Not a runtime or account-posture scanner. It's a change gate (plan / install), not a continuous scanner of deployed infrastructure or live cloud accounts. (A read-only account-posture watcher is on the roadmap.)
- Fewer raw IaC rules than Checkov/Trivy — depth on the transition + enforcement, not breadth of config checks.
When to reach for bumper
- You run an AI coding agent (Claude Code, Augment, Gemini CLI) and want it gated from destroying infra or installing malware — the case nothing else covers well.
- You want a deterministic apply gate that blocks on destruction/exposure, not just a linter that warns.
- You want one gate across infra + dependencies, in CI and in the agent loop, as a single static binary with no account.
- You care that dependency checks are lookup-not-upload and can be self-hosted (self-hosting.md).
It sits alongside Checkov/Trivy (broad config coverage), Dependabot (remediation PRs), and Socket (behavioral supply-chain) — not in place of them.