CodeQL is free on public repositories. On private repositories, GitHub requires GitHub Code Security, listed at $30 per active committer per month on a GitHub Team or Enterprise plan. If that budget isn't there, Skylos is a free, open-source check you can run on private code, on your machine and as a pull-request gate. It is not a replacement for CodeQL's data-flow analysis, and teams that can run both usually should.
GitHub and CodeQL facts checked on 21 Nov 2026 against GitHub's own pages, linked below. Skylos facts are for Skylos 5.7.0. Prices change; follow the links for the current terms.
Is CodeQL free for private repositories?
No. GitHub's code scanning docs say code scanning is available for "Public repositories on GitHub.com" and "Organization-owned repositories on GitHub Team, GitHub Enterprise Cloud, or GitHub Enterprise Server, with GitHub Code Security enabled." The same page says: "If you want to use code scanning on private repositories, you need a GitHub Code Security license."
GitHub's private repository troubleshooting page is blunter: "If you are on a GitHub Free or GitHub Pro plan, you can only use code scanning on repositories that are publicly available."
What that costs, from GitHub's pages on 21 Nov 2026:
| What you run | What GitHub requires | Source |
|---|---|---|
| CodeQL on a public repository | Nothing. "All public repositories have access to code scanning, secret scanning, and dependency review." | Advanced Security billing |
| CodeQL on a private repository | GitHub Code Security: $30 USD per active committer per month | github.com/security/plans |
| The plan underneath | Code Security is an add-on to GitHub Team ($4 USD per user per month) or Enterprise ($21 USD per user per month) | github.com/pricing |
| Secret scanning on a private repository | GitHub Secret Protection, a separate $19 USD per active committer per month | github.com/security/plans |
| AI Scan and agentic autofix | Code Security AI credits, "even in public repositories that don't require active-committer licenses" | Advanced Security billing |
| The CodeQL CLI on private code, outside a paid licence | Not allowed by the CodeQL terms (quoted below) | LICENSE.md |
How GitHub counts: "A committer is considered active if one of their commits has been pushed to the repository within the last 90 days, regardless of when it was originally authored," and "GitHub App bots are ignored" (billing docs, 21 Nov 2026). At list price, 20 active committers on private repositories come to 20 x $30 = $600 a month for Code Security, before the plan itself. Code scanning also runs in GitHub Actions, "with each workflow run consuming GitHub Actions minutes."
What the CodeQL CLI terms allow
The GitHub CodeQL Terms and Conditions (last changed 2 Feb 2026, read 21 Nov 2026) let anyone:
- Use the Software to perform academic research.
- Use the Software to demonstrate the Software.
- Test CodeQL queries that are released under an OSI-approved License to confirm that new versions of those queries continue to find the right vulnerabilities.
And, "only with an Open Source Codebase":
- Perform analysis on the Open Source Codebase.
- If the Open Source Codebase is hosted and maintained on GitHub.com, generate CodeQL databases for or during automated analysis, CI, or CD.
The restrictions exclude using it "in connection with any codebase that is not an Open Source Codebase (e.g., code in a private repo in GitHub)". They add: "if your use of the Software is under a paid customer license for GitHub Advanced Security, the restrictions with respect to automated analysis, CI, and CD and use in connection with non-Open Source Codebases do not apply." GitHub's CodeQL CLI page says the same in short: "The CodeQL CLI is free to use on public repositories."
That is also why this page has no CodeQL output: our test repository was private scratch code.
How do CodeQL and Skylos compare?
| CodeQL | Skylos | |
|---|---|---|
| Licence and cost | The CodeQL CLI is licensed under the GitHub CodeQL Terms and Conditions; the queries are open source. Free on public repositories. Private repositories need GitHub Code Security ($30 per active committer per month) on GitHub Team or Enterprise | Apache-2.0. "The CLI, the agent hooks and the PR gate in your own CI: free on any repository, private ones included, with no account" (skylos.dev/pricing, 21 Nov 2026). Skylos Cloud is optional |
| Where it runs | GitHub Actions on GitHub-hosted or self-hosted runners (default or advanced setup), or the CodeQL CLI in another CI with results uploaded to GitHub. Also GitHub Enterprise Server | Your laptop, inside Claude Code, Codex and Cursor through hooks, and any CI: skylos cicd init writes GitHub Actions, GitLab CI, Jenkins, Bitbucket Pipelines, Azure Pipelines or CircleCI config |
| Account needed | A GitHub repository. For a private one, an organization on GitHub Team or Enterprise with Code Security | None. Skylos Cloud is optional |
| Languages | C/C++, C#, Go, Java/Kotlin, JavaScript/TypeScript, Python, Ruby, Rust, Swift and GitHub Actions workflows. GitHub: "CodeQL does not support languages that are not listed above. This includes, but is not limited to, PHP, Scala, and others" | Named per check in the table under "What does Skylos add?". Security: Python, TypeScript/JavaScript, Java and PHP, partial for Go, Rust, Ruby, Dart, C#, Kotlin and Shell. No security checks for C++ or Swift |
| What it finds | Security and error queries run over a database built from your code, including data flow between functions and through object properties | Security flaws, secrets, dependency CVEs, quality issues, dead code, and checks aimed at coding agents: deleted or skipped tests, packages that do not exist, removed security controls |
| When it runs | Default setup: when enabled, on pushes to the default or a protected branch, on pull requests against them (not from forks), and weekly | Whenever you or your agent run it: after each agent edit (hooks), when the agent says it is done (skylos done), and on each pull request (the generated workflow) |
| PR gating | A ruleset with "Require code scanning results", with alert and security-severity thresholds per tool | --gate exits 1 on a failed gate. It always fails on a critical security issue, a secret or a critical AI-code issue; other limits go in [tool.skylos.gate], read from the base branch |
| Silencing a finding | Dismiss the alert with a reason, from the code annotation or the Security and quality tab. "It's dismissed in all branches" | # skylos: ignore[SKY-D211] reason on the line, or a baseline (skylos baseline .) |
| Fixes | Copilot Autofix suggests fixes for CodeQL alerts, with no Copilot subscription needed | Reports findings, with no fix suggestions. skylos clean removes unused Python imports and functions |
| Setup | Default setup: Settings, Advanced Security, "CodeQL analysis", Set up, Default. No workflow file | pip install skylos, then skylos cicd init --no-upload and commit the workflow it writes |
Sources for the CodeQL column, all read 21 Nov 2026: code scanning with CodeQL, setup types, default setup, merge protection, resolving alerts, AI features and data-flow analysis. Skylos column: README at v5.7.0.
What does CodeQL do better?
- Data flow across a whole program. CodeQL's docs describe global data flow as considering "the data flow within an entire program, by calculating data flow between functions and through object properties" (codeql.github.com). GitHub notes that "some complex security behaviors may only be detected on a complete build" (analysis time). That is the job it was built for, and Skylos's security rules are narrower.
- Security queries per language. GitHub's Python query list for the default and security-extended suites covers SQL injection, SSRF, path injection, unsafe deserialization, template injection, XXE, weak cryptography and more, with related CWEs.
- Languages Skylos has no security checks for. C/C++ and Swift are CodeQL languages. Skylos 5.7.0 has no security checks for either.
- Fixes inside GitHub. Copilot Autofix "is available for CodeQL analysis and does not require a GitHub Copilot subscription", and it is on by default for every repository using CodeQL (AI features). Skylos does not write security fixes.
- One switch for the organization. Default setup needs no workflow file, picks languages itself and can be applied across an organization. Alerts live in the Security and quality tab, and rulesets block merges by alert severity.
- Custom queries and models. The queries are open source, advanced setup can run extra query packs, and model packs extend CodeQL to frameworks it does not recognize.
- More from the same licence. Code Security also lists AI Scan for "languages and frameworks CodeQL doesn't cover" (AI credits), security campaigns and Dependabot auto-triage rules (github.com/security/plans).
- Copilot's own agent already runs it. GitHub's 28 Oct 2025 changelog says Copilot coding agent checks the code it writes with CodeQL, the GitHub Advisory Database and secret scanning, and that this does not "require a GitHub Advanced Security license or any additional configuration".
If your repositories are public, or your security team already pays for Code Security, keep CodeQL. That decision doesn't need this page.
What does Skylos add?
Skylos is a local-first, open-source check for code written with coding agents such as Claude Code, Codex and Cursor. It needs no account, and a scan calls no model, so its findings don't depend on a model's output.
| Check | Command | Languages |
|---|---|---|
| Deleted, skipped or weakened tests, loosened test settings and CI test steps (SKY-A110, SKY-A111, SKY-A112, SKY-A121) | skylos done --base main (runs your tests) | Python and JS/TS tests |
| Imports of packages that do not exist (SKY-D222) | skylos . -a | Python imports; requirements, package.json and go.mod entries |
| A security control removed by the change, such as a login decorator or CSRF check (SKY-L021) | skylos . -a --diff-base origin/main --diff origin/main | Python, TypeScript, JavaScript, Go, Java, PHP, Rust, Dart |
| Secrets | skylos . -a (or --secrets) | Source, config and text files |
| Dependency CVEs | skylos . -a (or --sca) | Exact versions in manifests and lockfiles, checked against OSV |
| Dead code | skylos . | Python, TypeScript/JavaScript, Java, Go, PHP and Dart; Rust experimental; others partial |
| Security flaws: SQL injection, SSRF, path traversal, command injection, unsafe deserialization | skylos . -a (or --danger) | Python, TypeScript/JavaScript, Java and PHP; partial for Go, Rust, Ruby, Dart, C#, Kotlin and Shell |
Sources: README and rule dictionary at v5.7.0. GitHub's Python query list, read 21 Nov 2026, has no query about whether an imported package exists on a registry or whether a branch deleted or skipped a test. CodeQL does list "CSRF protection weakened or disabled" for Python. SKY-L021 works differently: it compares each changed file with its earlier version and reports a control the change took away.
What did Skylos report on an agent-style branch?
We wrote a small Flask service with one endpoint on main, then a branch that does what a rushed agent might: it adds a search endpoint that builds SQL from a query parameter, and imports a helper package that does not exist on PyPI.
The branch's change (git diff main -- app.py):
@@ -1,7 +1,8 @@
"""Order lookup service."""
import sqlite3
-from flask import Flask, jsonify
+from flask import Flask, jsonify, request
+from sqlsafe_query_builder import paginate
app = Flask(__name__)
@@ -17,3 +18,13 @@ def get_order(order_id):
if row is None:
return jsonify({"error": "not found"}), 404
return jsonify({"id": row[0], "status": row[1]})
+
+
+@app.get("/orders/search")
+def search_orders():
+ customer = request.args.get("customer", "")
+ db = get_db()
+ rows = db.execute(
+ f"SELECT id, status FROM orders WHERE customer = '{customer}'"
+ ).fetchall()
+ return jsonify(paginate(rows, page=request.args.get("page", 1)))
Skylos on that branch:
$ skylos . -a --diff-base main --diff main --gate; echo "exit code: $?"
--diff-base: 1 changed files (full scan on changed, defs/refs-only on rest)
--diff: filtered to 2 changed line ranges from main
Quality Gate: FAILED
• 1 critical security issue
SKY-D211 app.py:27 Possible SQL injection: tainted or string-built query.
• 1 critical AI-code issue (SKY-D222 dependency hallucination)
SKY-D222 app.py:5 Hallucinated dependency 'sqlsafe_query_builder'. Package does not exist on PyPI.
Critical AI-code issues fail the gate by default. To allow them, set block_ai_defect_severities = [] under [tool.skylos.gate] in pyproject.toml.
exit code: 1
Skylos 5.7.0, 21 Nov 2026, macOS 15.3.2 on Apple silicon, Python 3.13.18. Scratch repository written for this page, branch agent/order-search against main, requirements.txt with flask==3.1.3. The banner, the excluded-folder list and eight dependency-inventory lines are cut. Nothing else is changed.
How to read it:
- The SQL injection is CodeQL's home ground. GitHub's Python list includes "SQL query built from user-controlled sources". We didn't run CodeQL (see the terms above), so we don't claim what it would report here. Skylos flagging it too is overlap, not the reason to add Skylos.
- The package that does not exist is the part CodeQL's published query list doesn't cover. Skylos checked the import against PyPI, found no such project, and failed the gate.
pip installwould fail on it today too. If someone registers the name later, the install would succeed. - Changed lines are what count.
--diff-base mainscans the changed files in full and reads the rest for definitions and references, and--diff mainkeeps the findings on lines the branch changed. Findings already onmaindon't fail the branch.
Where does Skylos fall short?
- Its security analysis is narrower than CodeQL's. Expect false positives, especially on code that builds SQL with query-builder libraries. Silence one with a reason,
# skylos: ignore[SKY-D211] customer is validated by the router, or accept existing findings withskylos baseline .and scan with--baseline. A removed security control (SKY-L021) can't be hidden by either. - Python has the deepest coverage. Several languages are partial, and C++ and Swift have no security checks.
- No security fixes. Skylos reports. CodeQL has Copilot Autofix.
- No organization view in the CLI. Skylos Cloud adds history and a dashboard; it is optional and priced on /pricing.
skylos donecovers Python and JS/TS tests. Other test runners can be run with atest_command, but the deleted and skipped test checks read Python and JS/TS test files.- Some checks use the network. Dependency and package checks send package names and versions to OSV and the public registries. Source stays on your machine.
- A failed job is not a blocked merge. Making the Skylos job required is a setting on your Git host. On GitHub, github.com/pricing lists repository rules on the Free plan for public repositories, so a private repository on Free can see the failure but can't require it.
- Benchmarks. Our public benchmark is self-reported for Skylos 5.3.0 on 43 open-source Python projects (1 Nov 2026), against Ruff, Pylint, Vulture, deadcode, CytoScnPy, Bandit and Semgrep CE. CodeQL was not part of it, and this page makes no accuracy comparison with CodeQL.
When should you use CodeQL instead?
- You need deep taint tracking across a large codebase. Data flow between functions and through object properties, over a database built from the whole program, is what CodeQL was built for.
- Your repositories are public. CodeQL, Copilot Autofix and SARIF uploads are free there, and default setup takes a few clicks.
- Your workflow lives in GitHub. Alerts in the Security and quality tab, merge rules by severity, security campaigns and org-wide enablement are all GitHub-native.
- Your security team already pays for GitHub Advanced Security. Code Security is already licensed. Add Skylos only if you want the agent checks.
- You ship C/C++, Swift or Kotlin and need security queries for them.
- You want to write your own queries in QL, or model an in-house framework for data flow.
When does Skylos fit?
- Private repositories without a Code Security budget. The CLI and the pull-request gate are free on any repository, with no account.
- You want the check before the push. It runs on a laptop and inside the agent's loop (
skylos agent install-hooksfor Claude Code,--codex,--cursor), as well as in CI after the push. - Your agents fail in ways a security query doesn't look for. Tests deleted or skipped to get to green, imports of packages that don't exist, a login decorator dropped in a refactor.
skylos done --base mainand the gate catch those. - Your CI is not GitHub Actions.
skylos cicd initalso writes GitLab CI, Jenkins, Bitbucket Pipelines, Azure Pipelines and CircleCI config. - Your code includes PHP. GitHub's docs list PHP as unsupported by CodeQL. Skylos 5.7.0 has PHP security and dead-code checks.
How do you run CodeQL and Skylos together?
They answer different questions, so many teams should run both on the same pull request.
- Turn on CodeQL default setup for public repositories, or private ones with Code Security: Settings, Advanced Security, "CodeQL analysis", Set up, Default.
- Add the Skylos gate. It needs no account:
pip install skylos
skylos cicd init --no-upload
git add .github/workflows/skylos.yml
git commit -m "Add Skylos pull request gate"
- Make both required. For CodeQL, a branch ruleset with "Require code scanning results" and your thresholds. For Skylos, mark its job as a required status check.
- Optional: one Security tab. If Code Security is on, the Skylos Action can upload SARIF so its findings show up as code scanning alerts next to CodeQL's. GitHub accepts SARIF uploads "for organization-owned repositories with GitHub Code Security enabled, and public repositories on GitHub.com" (CodeQL CLI page). The fragment below follows the Skylos 5.7.0 docs, pinned to the release commit:
permissions:
contents: read
security-events: write
steps:
- uses: actions/checkout@v4
- uses: duriantaco/skylos@87978e69adf91b3811e97b00c40aa39d7937cd94 # v5.7.0
with:
analysis: "security secrets sca ai-defects"
upload-sarif: "true"
sarif-category: "skylos"
- Check the agent before it opens the pull request.
skylos agent install-hookschecks each edit and blocks installs of packages that do not exist, andskylos done --base mainruns your tests and fails on deleted or skipped ones. - See where they overlap. If you already produce a CodeQL SARIF file under your licence,
skylos compare . --against codeql.sariflists the findings both tools report at the same location and those only one of them reports. It is advisory and does not change either gate.
To try it on your own branch:
pip install skylos
skylos . -a --diff-base origin/main --diff origin/main --gate
skylos done --base main
Changelog
- 21 Nov 2026: First version. GitHub pricing, Code Security availability, CodeQL CLI terms, setup, merge protection and Copilot Autofix facts read from GitHub's pages the same day. Skylos 5.7.0. One real Skylos run; CodeQL not run (licence terms).
Related
- Bandit vs CodeQL vs Semgrep for Python
- Semgrep vs Skylos
- GitHub Code Quality vs Skylos
- GitHub Copilot security review
- AI agent says the tests pass? Verify before merge
- Your AI agent tried to install a package that doesn't exist
- A free pull-request gate
References
All opened on 21 Nov 2026.
- GitHub security plans and pricing: Code Security $30 and Secret Protection $19 USD per active committer per month, feature availability by plan
- GitHub pricing: Team $4 and Enterprise $21 USD per user per month, repository rules by plan
- About code scanning: availability, licence for private repositories, Actions minutes
- Cannot enable CodeQL in a private repository
- GitHub Advanced Security licence billing: active committers, free use on public repositories, AI credits
- Code scanning with CodeQL: supported languages
- About setup types for code scanning and configuring default setup
- Code scanning analysis takes too long: "Analysis time is typically proportional to the amount of code being analyzed"
- Set code scanning merge protection
- Resolve code scanning alerts
- GitHub security and quality AI features: Copilot Autofix
- Python queries for CodeQL analysis
- About the CodeQL CLI and the GitHub CodeQL Terms and Conditions (last changed 2 Feb 2026)
- About data flow analysis (CodeQL docs)
- Copilot coding agent now automatically validates code security and quality (GitHub changelog, 28 Oct 2025)
- Semgrep pricing and Semgrep Community Edition
- Skylos README, rule dictionary, CLI output and SARIF and scanner comparison at v5.7.0