Feature

Automated Vulnerability Remediation via Pull Request

Auto-remediation in security means automatically generating, testing, and merging code changes that fix a known vulnerability — without a human authoring the patch or opening the PR. Sentinel auto-remediates a defined class of changes (dependency bumps, Terraform secrets), gates every PR on your existing CI pipeline, and never merges on a red build. Everything else gets escalated to the action inbox for human resolution.

How Sentinel Auto-Remediates

When Sentinel identifies a finding eligible for auto-remediation, it follows a fixed, auditable pipeline:

1

Classify & Confidence Score

Each finding is classified by change type (dependency bump, secret rotation, IaC drift). A per-PR confidence score is computed from EPSS score, CVSS severity, CISA KEV status, and change blast radius. Only findings above the configured confidence threshold (default: 0.85) proceed to auto-PR creation.

2

Open the PR

Sentinel creates a branch, authors the minimal change (version bump, secrets manager reference replacement), and opens a PR against your default branch. The PR body includes the CVE ID, EPSS score, CISA KEV status, confidence score, and a link to the Sentinel finding in your inbox.

3

CI Gate

Sentinel waits for your CI pipeline to complete. If CI passes, the PR is merged. If CI fails, the PR is left open, the finding is escalated to the action inbox, and the CI failure is attached as context. Sentinel never merges a red build.

4

Audit Log Entry

Every auto-remediation action — PR opened, CI result, merge or escalation — is written to an append-only audit log with timestamp, CVE ID, confidence score, and the committer identity (Sentinel bot). This log satisfies SOC 2 CC6.1/CC7.1, PCI-DSS 6.3, and NIST SI-2 evidence requirements for automated patch management.

What Gets Auto-Fixed

Sentinel auto-remediates two categories of changes where the fix is mechanical and the risk profile of the change is well-understood:

Category 1 — Dependency Bumps

Patch and minor version updates for npm, PyPI, Maven, Go modules, and Cargo dependencies. Sentinel checks that the target version is published, verifies the changelog for breaking changes using release notes, and applies the change. SemVer patch updates (e.g., 2.4.1 → 2.4.3) auto-merge on CI green. Minor updates (2.4.1 → 2.5.0) auto-PR but require human approval on merge. Major updates always escalate to the inbox.

Category 2 — Terraform Secrets

Hard-coded credentials and API keys in Terraform configuration are replaced with references to AWS Secrets Manager, HashiCorp Vault, or GCP Secret Manager (based on your cloud provider detection). The PR replaces the literal value with adata.aws_secretsmanager_secret_version reference and adds the corresponding data block. The original secret value is flagged for rotation in your secrets manager as a follow-up action.

What Gets Escalated (Not Auto-Fixed)

Auto-remediation has explicit scope limits. The following categories are always escalated to the human action inbox — Sentinel never auto-PRs these:

  • SAST findings — SQL injection, XSS, path traversal, authentication flaws. These require understanding business logic, data flows, and application context. A mechanical fix would be wrong as often as it was right.
  • Major version dependency upgrades — Major versions have intentional breaking changes. The upgrade may require API surface changes, configuration migration, or behavioral differences that only the service owner can evaluate.
  • IaC configuration findings (non-secrets) — Overly permissive IAM roles, public S3 buckets, unrestricted security groups. These changes can alter production behavior in ways that require human sign-off on blast radius.
  • Container base image upgrades — Base image updates often pull in OS-level changes. Sentinel flags these as HIGH priority in the inbox with the recommended new digest, but does not auto-apply them.
  • Confidence score below threshold — Any finding where Sentinel's confidence score is below 0.85 (configurable) is escalated regardless of change type. The threshold is tunable per repository.

Why Competitors' Auto-PRs Break Prod

Auto-remediation is not new. Dependabot has existed since 2019; Snyk has offered fix PRs since 2020. But field teams report two consistent failure modes:

Snyk Fix PR Regression Rate

In a 2023 field test by security engineering teams at a mid-size SaaS company [reported in the Snyk community forum, October 2023], 5.3% of Snyk fix PRs introduced regressions — build failures, behavioral changes, or test failures — that required manual reversion. The root cause was Snyk opening fix PRs without waiting for a full CI run on a branch before recommending merge.

Dependabot CI-Failure Merges

Dependabot creates PRs and, in auto-merge mode, can merge on status check passage even when some required checks are still pending or when the "required checks" configuration is incomplete [GitHub Docs: Dependabot auto-merge behavior, 2024]. Teams with partial CI configuration have reported Dependabot merging PRs that failed integration tests because unit tests passed and integration tests were not in the required checks list.

Both failure modes share the same root: the auto-remediation system does not have a strict definition of "CI passed" and does not wait for it unconditionally. Sentinel requires explicit CI pass status on all configured required checks before any auto-merge proceeds — and defaults to "all checks must pass," not "required checks must pass."

Sentinel's Safety Model

MechanismWhat it prevents
CI gate (all checks, not just required)Regressions from partial CI coverage configurations
Confidence score threshold (default 0.85)Auto-PRs on ambiguous or high-blast-radius changes
SemVer patch-only auto-mergeBreaking API changes sneaking through minor/major bumps
Append-only audit logUntracked automated changes in production; compliance gaps
Human-in-the-loop for merge escalationAuto-merge of changes where Sentinel confidence is below threshold
Per-repository routing rulesAuto-remediating in repositories where it is not appropriate (e.g., payments-api)

Auto-remediation is not "set and forget." It is "automate the safe class of changes and route everything else to a human with full context." The goal is not to remove humans from the loop — it is to reserve human attention for decisions that actually require it.

Frequently Asked Questions

What types of vulnerabilities can be auto-remediated?

Sentinel auto-remediates dependency version bumps (patch and minor updates for npm, PyPI, Maven, Go modules, Cargo) and Terraform hard-coded secrets(replacing literal credentials with secrets manager references). All other finding types — SAST, container images, IaC configuration, major version upgrades — are escalated to the human inbox with full context and remediation hints.

Will auto-remediation break my builds?

No, by design. Sentinel never merges a PR where CI has not passed on all configured checks.If CI fails, the PR stays open, the finding is escalated to the inbox, and the CI failure output is attached as context for the human reviewer. The only way a Sentinel auto-PR could break your build is if your CI pipeline has gaps that would allow a broken build to pass — in which case the CI pipeline, not the auto-remediation, is the issue.

How does Sentinel decide what to auto-merge vs escalate?

Three factors determine escalation:

  • Change type: Only dependency patch/minor bumps and Terraform secrets are eligible for auto-merge.
  • Confidence score: Findings must score above 0.85 (EPSS + CVSS + KEV status + blast radius) to auto-proceed. The threshold is configurable per repository.
  • CI result: All CI checks must pass. Any failure routes to escalation.

Anything outside these bounds — wrong change type, low confidence, CI failure — goes to the action inbox with the auto-PR left open for human review and merge decision.

Is auto-remediation safe for production?

Yes, with the caveat that "safe" depends on your CI pipeline coverage. Sentinel auto-merges only when CI passes on all checks. If your CI pipeline tests production behavior thoroughly (unit tests, integration tests, and contract tests), then a Sentinel auto-merge is as safe as any human-authored patch PR that also passes CI. For repositories where any automated merge is unacceptable — e.g., a payments service with a manual deployment gate — you can configure Sentinel to open PRs without auto-merge, so a human still controls the merge button.

Private beta · 2026

Stop writing vulnerability fix PRs manually.

Sentinel opens them, runs CI, and merges on green. Join the waitlist — 5-minute setup, no credit card required.

Private betaNo credit cardSOC 2 aligned