Skip to main content
Pipefort runs deterministic checks per scan across several surfaces:
  • 34 GitHub Actions workflow checks parse .github/workflows/*.yml (one, forbidden-uses, is driven by your .pipefort.yml policy).
  • 7 online supply-chain audits verify the integrity of pinned actions (known-vulnerable, impostor-commit, ref/version-mismatch, typosquat, archived-action, stale-action-ref, ref-confusion) — run automatically when a GitHub token is available, forced with --audit-pins, disabled with --offline.
  • 17 GitHub repository-configuration checks call the GitHub API for branch protection, default workflow permissions, secret scanning, and Dependabot.
  • 11 GitLab CI workflow checks parse .gitlab-ci.yml and .gitlab-ci/**/*.yml.
  • 5 GitLab project-configuration checks call the GitLab API for protected branches, merge policy, public-pipeline visibility, and approvals.
Workflow checks cover all ten OWASP CI/CD Top 10 categories (CICD-SEC-1 through CICD-SEC-10) on both platforms.

GitLab CI workflow checks

The GitLab rule IDs are parallel to the GitHub ones — same OWASP category, different IDs (e.g. cicd-sec-1-gl-mr-target for the merge-request-target analog of cicd-sec-1-ppe-checkout). Findings tagged -gl- only fire on .gitlab-ci.yml / .gitlab-ci/*.yml. The two structural rules best-prac-1-pipe-to-shell and cicd-sec-9-download-without-checksum share their ID across both platforms.

GitLab project-settings checks

These read the GitLab project configuration over the API (CLI: --git at a GitLab host with --gitlab-token; the web app uses its connected GitLab token) and surface under the Repository-configuration group. The two auto-fixable settings (merge-without-pipeline, public-pipelines) are remediated by the CLI’s --fix-settings-gl flag.

v1 limitations

  • include: graph traversal is not performed — the scanner inspects only the literal .gitlab-ci.yml and .gitlab-ci/*.yml contents, not files referenced from another project or URL.
  • Approval-rule auditing (cicd-sec-1-gl-no-approvals) needs the GitLab Premium approvals API; on Free tier it is skipped rather than firing.
Rules carry framework tags so they can be filtered by the standard they serve: OWASP Top 10 CI/CD Security Risks and the SLSA v1.2 specification (Build and Source tracks). A single rule can belong to multiple frameworks.

Confidence and personas

Every finding carries a confidence (HIGH/MEDIUM/LOW) alongside its severity: severity says how bad the issue would be, confidence says how sure the check is that the finding is real. Deterministic checks (a missing permissions: block either exists or doesn’t) are HIGH; heuristic ones (typosquat edit distance, secret-name patterns) are MEDIUM. Filter with the CLI’s --min-confidence flag; the web app badges medium/low-confidence findings and offers a “High confidence only” toggle. Rules are also tiered into personas for noise control: regular (the default — high-signal security checks), pedantic (adds hygiene nits like missing timeouts and job-level continue-on-error), and auditor (adds everything, e.g. self-hosted-runner usage). Select with the CLI’s --persona flag. The web app runs all tiers and relies on rule settings for per-rule control.

Workflow file checks

Online pinned-action audits

These audit the integrity of pinned actions and need network access. They run automatically whenever a GitHub token is supplied — --github-token, $GITHUB_TOKEN, or $GH_TOKEN on the CLI, and the web app’s scan always runs them with its installation token. Without a token the scan stays offline unless you force the pass with --audit-pins (subject to GitHub’s low anonymous rate limit); --offline disables it unconditionally. Typosquat matching is itself offline but ships in the same pass.

Repository configuration checks

These read GitHub-side settings and surface under a “Repository configuration” group in the UI (CLI: <repository settings> file label). They need the expanded GitHub App permissions described in GitHub App permissions; the CLI needs --github-token (or $GITHUB_TOKEN, or gh auth token).

Branch protection (CICD-SEC-1)

Actions runtime (CICD-SEC-4, CICD-SEC-5)

Dependency hygiene (CICD-SEC-3)

Credential hygiene (CICD-SEC-6)

Auto-fix on repo-settings rules is powered by the CLI’s --fix-settings flag and the web app’s per-finding Fix button — see Auto-fix.

Rulesets

The CLI’s --ruleset flag (and the web app’s ruleset selector) controls which checks contribute to the final list. Filtering is by framework membership, not by category prefix — see SLSA framework overview for a rule-by-rule mapping.
  • all (default) — every check listed above.
  • owasp — every rule tagged with the OWASP framework.
  • slsa — every rule tagged with any SLSA v1.2 framework (Build or Source).
  • slsa-build-l1 / slsa-build-l2 / slsa-build-l3 — rules for that specific Build level (and only that level — to find every rule a repo needs to satisfy L3, run with slsa and look at the heatmap).
  • slsa-source-l2 / slsa-source-l3 / slsa-source-l4 — rules for that specific Source level. (L1 is “Version Controlled” — trivially satisfied for any GitHub repo.)

Enabling and disabling individual rules

The web app lets you toggle any individual rule on or off without changing the ruleset — per user, with optional per-repository overrides. See Rule settings for the model and the UI. The CLI’s filtering is limited to the coarser --ruleset choice above; multi-tenant preferences depend on the database and are web-only.

How the checks run

Each workflow check parses a workflow file and returns findings. The CLI and the web app run the exact same checks, so they always agree for the same file. Repository-configuration checks are a separate pass: Pipefort reads the relevant GitHub settings over the API and produces findings from them. These carry a synthetic file label (<repository settings>) with no line number, so they render apart from per-file findings. See Auto-fix for which workflow categories the CLI’s --fix flag rewrites. Repository-configuration findings have no auto-fix — they’re flagged for manual remediation via GitHub’s UI.