Overview
Per semgrep.dev/docs/getting-started/quickstart (opens in new window):
Semgrep is a fast pattern-based static analyzer covering 30+ languages with a registry of community + paid rulesets and straightforward YAML rule authoring. The semgrep ci subcommand adds CI-aware features (baseline-diff, organization policies, metrics).
Per semgrep.dev/docs/cli-reference (opens in new window):
"
semgrep scan: Local scans without account requirement; doesn't return failing codes by default.
semgrep ci: Pipeline execution with organization policies, diff-aware scanning, returns failing codes on findings."
When to use
Step 1 - Install
Per sg-quick (opens in new window):
# macOS
brew install semgrep
# Linux/macOS
pipx install semgrep
# or
uv tool install semgrep
# Windows (PowerShell)
pipx install semgrep
# Docker (CI-friendly)
docker pull semgrep/semgrepStep 2 - First scan
semgrep scan --config autoPer sg-cli (opens in new window), --config auto "Auto-fetch rules from registry based on project." Specific rulesets:
semgrep scan --config p/owasp-top-ten # OWASP Top 10
semgrep scan --config p/default # broad community ruleset
semgrep scan --config p/python p/javascript # multipleStep 3 - Custom rule authoring
A minimal Semgrep rule in .semgrep.yml:
rules:
- id: hardcoded-jwt-secret
pattern: jwt.sign($PAYLOAD, "...")
message: Hardcoded JWT secret detected
languages: [javascript, typescript]
severity: ERROR
metadata:
cwe: "CWE-798: Use of Hard-coded Credentials"Pattern operators: pattern, pattern-either, pattern-not, metavariable-pattern, pattern-inside. Validate rule syntax:
semgrep validate --config .semgrep.yml(Per sg-cli (opens in new window) subcommand list.)
Step 4 - CI integration with baseline diff
Per sg-cli (opens in new window): "--baseline-commit=VAL - Show only findings not in specified commit". Diff-aware mode fails only NEW findings on the PR; pre-existing findings are tracked but don't block - critical for legacy adoption.
jobs:
semgrep:
runs-on: ubuntu-latest
container:
image: semgrep/semgrep
steps:
- uses: actions/checkout@v5
- run: semgrep ci --baseline-ref=main --sarif --output=semgrep.sarif
- uses: github/codeql-action/upload-sarif@v3
if: always()
with: { sarif_file: semgrep.sarif }Step 5 - False-positive triage (MANDATORY)
Suppression mechanisms in priority order:
| Mechanism | Example | When to use |
|---|---|---|
Per-line nosemgrep comment | # nosemgrep: hardcoded-password | Justified single-line exception |
nosemgrep block | # nosemgrep: rule-id above a code block | Multi-line exception |
paths.exclude in config | exclude: ["**/*_pb.go"] | Generated files / vendored code |
| Baseline ref | --baseline-commit=main (Step 4) | Legacy debt; ratchet |
| Organization-level rule disable | Semgrep AppSec Platform UI | Team-wide policy |
Justification template (mandatory in code):
# nosemgrep: hardcoded-password
# Reason: Test fixture; password never reaches production runtime
# Reviewer: alice@example.com (2026-05-15)
# Expires: 2026-12-15
TEST_PASSWORD = "test-only-password-do-not-deploy"Per sg-cli (opens in new window) severity filter for triage workflow:
semgrep scan --severity ERROR --json # only critical findingsCadence: every quarter, audit nosemgrep suppressions for staleness. Expired ones removed; persistent ones reviewed for escalation.
Step 6 - CLI reference (formats, flags, exit codes)
Gate on exit 0 = clean, 1 = findings, 2 = fatal error. The full output-format flags, performance-tuning flags, and complete exit-code table are in references/cli-reference.md.
Anti-patterns
| Anti-pattern | Why it fails | Fix |
|---|---|---|
--config=auto everywhere | Rulesets drift; no ownership | Pin specific rulesets (Step 2) |
nosemgrep without justification | Becomes invisible debt | Required justification template (Step 5) |
| No baseline ref | Every legacy finding blocks; team disables | --baseline-commit=main (Step 4) |
semgrep scan in CI | Doesn't return failing exit code by default | Use semgrep ci (Step 1 quote) |
Mix --severity ERROR with --baseline-commit poorly | Can mask real new findings | Severity filter at output stage, not scan stage |
Limitations
References
Full flag, output-format, and exit-code detail for semgrep-rules. The SKILL.md spine keeps the common commands; this file holds the exhaustive tables. All entries per semgrep.dev/docs/cli-reference (opens in new window).
Output formats
| Flag | Purpose |
|---|---|
--json | Semgrep JSON format (for multi-scanner triage) |
--sarif | SARIF format (GitHub Code Scanning upload) |
--gitlab-sast | GitLab SAST format (GitLab Security Dashboard) |
--junit-xml | JUnit XML (test reporters) |
--text | Default human-readable |
--output VAL | Write to file or URL |
Performance flags
semgrep scan -j 8 --timeout 10 --max-target-bytes 5000000Exit codes
| Code | Meaning |
|---|---|
| 0 | Success, no issues |
| 1 | Issues detected (with --error flag) |
| 2 | Fatal error |
| 3 | Invalid syntax in scanned language |
| 4 | Invalid pattern in rule |
| 5 | Invalid YAML configuration |
| 7 | Invalid rule in configuration |
| 8 | Unsupported language specified |
| 13 | Invalid API key |
The spine gates on 0 (pass) / 1 (findings) / 2 (fatal); the rest signal config or rule authoring errors.