Skip to main content

Watch and follow along

Skylos video tutorials

Choose a task, from a local scan to reviewing a team's pull requests. Every video has English captions and a text transcript.

For a complete setup-to-PR walkthrough, use the written tutorial. The CLI documentation covers current commands and options.

These recorded walkthroughs can show earlier versions. Notes beside the videos highlight current setup, upload and policy details. For current plans, allowances and credit costs, use the pricing guide.

Start with the CLI

Install Skylos, choose your checks and review an agent's change.

Install Skylos and read a local dead-code scan.

Current guide: The recording uses Skylos 5.1.1. Current releases require Python 3.11 or later.

Read the transcript

Skylos CLI · Chapter 1: Install and first scan One package, one command. Find the code nobody uses, without sending anything anywhere.

Step 1: Install Skylos One package from PyPI, Python 3.10 or newer. No account and no sign-up.

Step 2: Check the version This tutorial uses Skylos 5.1.1.

Step 3: Run your first scan skylos . finds dead code: unused functions, imports, classes and files. It runs on your machine; nothing is uploaded.

Recap: pip install skylos · skylos . finds dead code, locally · Each finding shows its confidence and evidence · The footer tells you what to run next Next: Chapter 2 · Every check in one scan

Run security, secrets, quality and dependency checks, then narrow the results.

Current guide: Dependency checks can send package names and versions to public registries and OSV.

Read the transcript

Skylos CLI · Chapter 2: Every check in one scan Security, secrets, quality, AI-code mistakes and vulnerable dependencies, with one flag.

Step 1: Turn on every check -a adds security flow, secrets, quality, AI-code defects and dependency vulnerabilities to the dead-code scan.

Step 2: One line per finding --format concise prints file:line, rule and message, ready for scripts. --category narrows what is listed.

Step 3: Run exactly the rules you want --select takes rule IDs and turns on the checks they need. Every rule is listed in the dictionary in the docs.

Recap: skylos . -a runs every check · --format concise for one line per finding · --category and --select narrow the results · Findings are grouped by Defect, Review and Advice Next: Chapter 3 · Check an AI agent's change

Compare a branch, run completion checks and review what changed.

Current guide: Some completion findings are advisory by default, including weakened assertions and undeclared dependencies. Use the current completion policy to decide what blocks.

Read the transcript

Skylos CLI · Chapter 3: Check an AI agent's change An agent says it is done. Five commands tell you what it really changed.

Step 1: Look only at what the branch changed --diff main reports findings on changed lines only, so old issues don't bury the new one.

Step 2: Ask Skylos whether it's really done skylos done runs your tests itself and blocks deleted or skipped tests, loosened settings, secrets and made-up packages.

Step 3: See what changed, riskiest first The change map lists every changed function, what it now calls, who calls it, and what to read first.

Step 4: Verify the changed code skylos verify gives a pass or fail for AI-code mistakes, high-severity security issues and secrets.

Step 5: Know what needs a human approval Risk review lists the changes a person must approve: weakened tests, removed auth, new dependencies, weakened CI.

Recap: skylos . -a --diff main · skylos done --base main · skylos done map --base main · skylos verify . --diff main · skylos cicd risk-review --base main Next: Chapter 4 · Gate locally and in CI

Check recorded dependency versions, review licences and export an SBOM.
Read the transcript

Skylos CLI · Chapter 5: Supply chain Know which dependencies are vulnerable, which licenses you ship, and hand over an SBOM.

Step 1: Find vulnerable dependencies --sca checks your recorded versions against the OSV.dev advisory database and tells you which version fixes them.

Step 2: Set a license policy List the licenses you allow, or the ones you deny, in [tool.skylos].

Step 3: See what breaks the policy Licenses come from installed package metadata, offline. Skylos never guesses: unknown stays unknown.

Step 4: Export an SBOM CycloneDX 1.6 by default, or SPDX 2.3 with --format spdx-json. Both carry the declared licenses.

Step 5: Every package, version and license Hand the file to customers, auditors or your security team.

Recap: skylos . --sca · license_allow or license_deny in [tool.skylos] · skylos sbom . -o sbom.cdx.json · --format spdx-json for SPDX 2.3 Next: Chapter 6 · Clean up

Connect Claude Code, Codex or Cursor hooks and review their feedback.

Current guide: Local hooks have coverage limits and can fail open. Use a required CI check for merge enforcement.

Read the transcript

Skylos CLI · Chapter 8: Agent guardrails Stop a coding agent before it installs a fake package, reads a secret or skips your checks.

Step 1: How the hooks work Claude Code, Codex and Cursor ask Skylos before risky actions. Skylos answers allow or deny, with a reason the agent reads.

Step 2: Install the hooks One command adds Skylos to Claude Code's settings. Add --codex or --cursor for those agents.

Step 3: What the agent sends This is the event Claude Code sends before it runs a shell command. Here the agent is about to install "reqeusts".

Step 4: Block fake and typo-squatted packages The package does not exist on PyPI, so anyone could register it. Skylos denies the install and tells the agent how to fix it.

Step 5: Keep secrets out of the model app/payments.py holds a Stripe key. Reading it would send the key to the model provider, so Skylos blocks the read.

Step 6: No skipping your git hooks Agents reach for --no-verify when a pre-commit hook fails. Skylos blocks it and tells the agent to fix what the hook reports.

Step 7: See what the hooks stopped agent recap counts every block this week. The log stays on your machine.

Recap: skylos agent install-hooks · Fake and typo-squatted installs are denied · Files with secrets are never sent to the model · --no-verify is blocked when you have git hooks · skylos agent recap Next: Chapter 9 · AI assistant features

Work with your team

Connect a workspace, set policy and review changes together.

Connect a repository and open its uploaded scan in Cloud.

Current guide: Cloud uploads can include findings, code excerpts and provenance metadata. Review the current upload options and privacy guide before connecting. Read the privacy guide.

Read the transcript

Skylos Cloud · Chapter 11: Connect Skylos Cloud One command signs you in, uploads the findings and gives you a link.

Step 1: Local first, Cloud when you choose skylos . never sends anything. skylos cloud uploads the findings, not your code, so your team sees the same result.

Step 2: Connect and upload in one command The first run opens your browser to pick the project. Nothing is scanned or uploaded until you connect.

Step 3: Pick the project once Skylos recognizes the repository. Connecting issues a revocable project credential; it uploads no code.

Step 4: See it in Skylos Cloud The link opens the repository. A laptop upload shows the result but doesn't set the official verdict; uploads from CI do.

Recap: skylos cloud · Pick the project in the browser once · Findings are uploaded, never your code · Laptop uploads are unverified; CI sets the verdict · skylos whoami shows the connected project Next: Chapter 12 · Your workspace

Explore repositories, branches, gate history and trends.
Read the transcript

Skylos Cloud · Chapter 12: Your workspace Every repository, branch and gate result in one place.

Step 1: Start at Home Home shows how many repositories pass, one row per repository, with why it fails and what blocks it.

Step 2: See your coverage Which repositories are scanned, which fail the gate, and how many critical or high issues are open.

Step 3: Switch branches Every branch and pull request keeps its own gate result. Here the agent's branch fails on one critical issue.

Step 4: Every result, in order Gate history lists each result on the default branch, so you can see when it started failing and why.

Step 5: Watch the trend Trends shows issues, scans and pull request checks over 7, 30 or 90 days, for one branch or all of them.

Step 6: Rename your workspace Owners rename the workspace in Settings. The change is recorded in the audit log.

Step 7: Add another workspace Keep separate teams apart with their own repositories, policy and members. New workspaces start on the Free plan.

Recap: Home: one row per repository · Which repos are scanned · Branch picker and gate history · Trends over 7, 30 or 90 days · Rename the workspace in Settings · New workspace from the sidebar Next: Chapter 13 · Team policy, end to end

Edit gate conditions, preview their impact and review policy history.
Read the transcript

Skylos Cloud · Chapter 13: Setting team policy Decide once what fails a pull request. Every repository, every laptop and every pull request then gets the same answer.

Step 1: Open Policy Policy lives under Governance in the sidebar. Admins and owners can change it; everyone else can read it.

Step 2: One page, five parts Quality gate, Coding agents, Supply chain, Custom rules and History. One Save covers them all.

Step 3: Start from the default A new workspace blocks any new issue. Critical security issues always count, and exposed secrets are caught by that same rule.

Step 4: Edit the conditions Start from a preset (Recommended, Stricter on AI code or Strict) or set the rules yourself. Nothing is saved until you press Save.

Step 5: Allow a few, block what matters For older repositories: allow a handful of new medium and low issues, and keep critical and high at zero.

Step 6: Always block exposed secrets Tick this so secrets stay blocked even if someone loosens the limits later.

Step 7: Say why Skylos lists every change before you save. The reason is kept with this version in Policy history.

Step 8: Preview before you save Preview re-checks the latest scan of every repository against the new rules. Nothing is saved yet.

Step 9: Save it Each save becomes a new version. Repositories that inherit the policy use it on their next upload.

Step 10: Stop repositories loosening it Turn this on and a repository can make its own policy stricter, never looser.

Step 11: Rules for code an AI agent wrote Turn on agent rules, then choose Block, Warn or Off for each signal, such as weakened tests or a made-up package.

Step 12: Supply chain Block malware, hold brand-new package releases, and name your internal packages so look-alikes on public registries are blocked.

Step 13: Your own rules Custom rules add checks for patterns the built-in rules miss. They run in the CLI and on every pull request.

Step 14: Every change is a version Policy history shows who saved each version, when and why. Restoring an old one saves it again as a new version.

Step 15: Check a branch from your laptop skylos cloud --gate scans, uploads the findings and exits 1 when the team policy fails the branch.

Step 16: See why in Skylos Cloud The link opens the branch: the condition that failed, its limit and the count in this scan. A lead can still choose Merge anyway, with a recorded reason.

Recap Policy → Edit conditions · Reason for change, then Preview impact · Apply to every repository (stricter only) · Coding agents, Supply chain, Custom rules · History → Restore this version · skylos cloud --gate

Review agent checks, protected paths and team approval rules.
Read the transcript

Skylos Cloud · Chapter 14: Rules for AI agents Hold code an AI coding agent wrote to your rules before anyone merges it.

Step 1: Check the agent, then record the result skylos done checks the agent's change. skylos cloud uploads the scan with that result, and Skylos Cloud applies your agent rules.

Step 2: Check the agent's work Claude wrote this branch. It skipped a test and added a package that doesn't exist on PyPI.

Step 3: Upload it with the result skylos cloud --done-receipt sends the scan and the done result together. The gate fails on the agent's change.

Step 4: Coding agents, in one place AI, then Coding agents: how many pull requests agents opened, how many failed a check, and why.

Step 5: See which change failed Agent checks lists every checked change with its result. Open the one that failed.

Step 6: Why it failed The check names the rule that blocked, and links to the scan with the lines.

Step 7: The branch fails the gate Failed agent checks and new issues in files the agent wrote both count against the branch.

Step 8: Rules for agent pull requests Policy, then Coding agents. On pull requests with agent-written code, block new issues at the severity you choose.

Step 9: Block or warn, per risk Weakened tests, removed security controls, made-up packages and more: each one blocks, warns or is off.

Step 10: Protect the paths that matter An agent's change to a protected path waits for the team you pick to approve it on the Pull requests board.

Step 11: Bot approvals don't count alone When an AI reviewer approves a risky change, the Skylos approval guard holds it until a person approves too.

Step 12: Checks before an agent says it's done Agent guardrails set what skylos done and the hooks on developer machines enforce, for every repository at once.

Recap: skylos done checks the agent's change · Coding agents: what agents wrote and what failed · Agent rules block by severity and by risk · Protected paths and bot approvals wait for a person · Agent guardrails apply to every repository Next: Chapter 15 · Custom rules

Create a workspace rule and keep rule packs in the repository.

Current guide: Rule limits depend on your current plan. Use the pricing guide for current allowances. View pricing.

Read the transcript

Skylos Cloud · Chapter 15: Custom rules Write a check for the pattern your team bans. Every connected scan runs it.

Step 1: Open Custom rules Custom rules live under Policy. Workspace runs up to 50; Free runs 3.

Step 2: Start from a template "No Raw SQL Strings" flags database calls built from strings. Give it your own rule ID; every finding carries it.

Step 3: Create it The rule runs on future scans from every connected project.

Step 4: It runs in every connected scan On the agent's branch, a local scan catches string-built SQL with your rule. Connected scans load your workspace's rules.

Step 5: Change it any time Edit raises this rule to Critical. Earlier findings keep the rule ID.

Step 6: Duplicate to adapt Duplicate makes a copy that stays off until you turn it on, so you can adapt it safely.

Step 7: Keep rules in the repository too skylos rules init writes a rule pack into the repository. Quality scans apply it; in pull request CI, the copy on the base branch applies.

Step 8: Rules are plain YAML Each rule has an ID, a severity, a message and a pattern. Edit it, then check it with skylos rules validate.

Recap: Policy → Custom rules → Create a rule · Start from a template, use your own rule ID · Connected scans run your rules · Edit or duplicate any rule · skylos rules init: rule packs in the repo · Workspace runs 50 rules; Free runs 3 Next: Chapter 16 · Pull requests and merge blocking

Read a gate verdict, inspect its conditions and compare scans.

Current guide: Use current pricing and the in-app confirmation for feature availability and credit costs. View pricing.

Read the transcript

Skylos Cloud · Chapter 17: Read your results Every upload gets a quality gate, the conditions behind it, and the issues that caused it.

Step 1: One result per upload Each upload from skylos cloud or CI gets a pass or fail, the conditions that decided it, and the issues behind them.

Step 2: Start with the verdict The scan page opens on the quality gate: pass or fail for this branch and commit, and what to fix first.

Step 3: See which conditions failed Each condition shows its limit and what this scan found. Failed conditions come first; the ones that passed fold away below.

Step 4: New code, or already there Every issue says whether this change brought it in. New ones that fail a condition are marked Blocks merge; older ones are Not new.

Step 5: Open an issue The issue shows the exact lines, why it blocks the merge, similar issues, and Not fixing for a reviewed decision.

Step 6: Check the health of the code Security and correctness risk, maintainability and dead code, with what each scan covered.

Step 7: Compare two scans Pick an older and a newer scan to see what is new, fixed and unchanged. A comparison costs 2 credits.

Recap: Quality gate: pass or fail per branch · Conditions: limit against this scan · New issues block; Not new ones don't · Open an issue for lines, evidence and Not fixing · Health panel and scan comparison Next: Chapter 18 · Not fixing

Review false positives, accepted risks and merge overrides.
Read the transcript

Skylos Cloud · Chapter 18: Not fixing Decide what you won't fix, and why. Every decision keeps a reason, a person and a date.

Step 1: Start from the issue Open any issue and choose Not fixing this.

Step 2: Pick the kind of decision False positive hides it for good. Accepted risk hides it until a date. Merge anyway lets one failed scan through.

Step 3: False positive: say why A reason is required. With one owner the decision applies straight away; with a team, a second person approves it.

Step 4: Accepted risk: until a date Accept a dependency advisory for 30 days. On that date it shows up again in new scans.

Step 5: Merge anyway, on the record A lead can let one failed scan through with a reason. The issues stay open; only this scan's gate passes.

Step 6: Every decision in one place Not fixing lists each decision with who made it, when, why and until when, and exports them all for auditors.

Step 7: Rules for every repository Hide one issue in every repository at once. Secrets and high or critical security issues can't be hidden this way.

Recap: Issue → Not fixing this · False positive: hidden, with a reason · Accepted risk: hidden until a date · Merge anyway: one scan passes, issues stay open · Not fixing page and CSV export Next: Chapter 19 · Evidence and compliance

Find audit logs, access reviews and evidence reports.
Read the transcript

Skylos Cloud · Chapter 19: Evidence and compliance Every upload, decision and policy change is recorded, ready for an auditor.

Step 1: Start at Audit evidence One page for security evidence, change history, access review, decisions and policy history.

Step 2: The audit log Every upload, merge past a failed gate, Not fixing decision and policy change, with who did it and when.

Step 3: Review who has access Members, roles and each person's latest activity, so stale owners and shared tokens stand out.

Step 4: Compliance reports from verified scans Reports use scans uploaded from CI, because laptop uploads aren't verified. The check is free; a report is charged once.

Recap: Audit evidence: one page for auditors · Audit log with CSV, JSON and NDJSON · Access review: roles and latest activity · Compliance reports from CI-verified scans Next: Chapter 20 · Team, integrations and plans

Find who needs to act and review protected changes.
Read the transcript

Skylos Cloud · Chapter 21: Pull requests board See whose turn every pull request is, and approve risky changes with a tick.

Step 1: Every pull request has one owner right now Skylos sorts each pull request by who has to act next: the author, a team, or nobody.

Step 2: An AI agent touched payment code Claude wrote pull request #12 and it changes the protected Payments path, so the Skylos check waits for the Leads team.

Step 3: One board for every pull request Pull requests, then Board. Each column says whose turn it is. List and Board show the same pull requests.

Step 4: Only what's waiting on you Waiting on you keeps the cards you can act on: approvals your team owns, and fixes on your own pull requests.

Step 5: Open a card The panel says why the pull request is here and lists each thing that needs your approval.

Step 6: Tick it, add a note, approve The note is required and goes in the audit log. You can only approve paths your team owns, and never your own change.

Step 7: Approved, and recorded Skylos saves who approved what, for which commit, and re-runs the GitHub check.

Step 8: The card moves by itself Nothing else blocks #12, so it moves to Ready to merge.

Step 9: GitHub turns green The Skylos check now passes on GitHub. The merge button is the team's normal one.

Step 10: Or send it back For #13, write what needs to change and choose Request changes instead.

Step 11: It's the author's turn again #13 moves to Needs fixes with your note on it. When the author pushes a fix, Skylos checks the new commit.

Step 12: The author sees it on GitHub too Skylos posts your note as a comment on the pull request, so the author doesn't need to open Skylos.

Recap: One board: whose turn each pull request is · Waiting on you: only what you can act on · Tick what you approve; the note goes in the audit log · Approving re-runs the GitHub check · Request changes posts your note on GitHub Next: Chapter 22 · Compliance reports

Choose verified scans, review evidence and complete management answers.

Current guide: Evidence reports support an assessment. They are not a certification or an auditor's opinion.

Read the transcript

Skylos Cloud · Chapter 22: SOC 2 evidence report Turn your scans, fixes and pull requests into evidence an auditor can test.

Step 1: From scans to an auditor's evidence Trusted CI scans and your pull requests become a SOC 2 pre-assessment. You add what only you can write, and the auditor samples from the files.

Step 2: Check which projects count Compliance reports, then Step 1. Only trusted CI scans of the default branch count. An upload from a laptop doesn't.

Step 3: Pick SOC 2 Each framework has its own report. It's charged once, and every format downloads free.

Step 4: Choose projects and the audit period Pick the repositories in scope and the dates your auditor will test, up to 12 months.

Step 5: Signed and ready Skylos signs the report and stores it. The CSV files are the populations an auditor samples from.

Step 6: Start with the result Policy met, fixes due: one critical dependency advisory, still inside its fix-by date. It's Skylos's policy result, not a SOC 2 opinion.

Step 7: Management's assertion Every SOC 2 report opens with one, signed by an officer. Skylos never writes it for you; it gives you a draft to adapt.

Step 8: Controls, tested like an auditor would Each control has a test, a population and a result in SOC 2 wording: no deviations, or x of y.

Step 9: It says what it couldn't test GitHub didn't report reviews for this merge, so the approval test says not evidenced instead of passing.

Step 10: Readiness, criterion by criterion A criterion is Implemented only when every Skylos test for it passed. One unproven test makes it Partially implemented.

Step 11: Deviations, with your response Each deviation is listed with a place for management's response. Write it in SOC 2 answers and it prints here next time.

Step 12: Send auditors the external summary The same tests and results, without code locations, issue text or people's emails.

Step 13: Auditors can check it's genuine The verification page confirms the copy is unchanged and signed by Skylos, without showing any of its content.

Step 14: Fill in what only you can write SOC 2 answers: the system description, approved control wording, readiness owners and deviation responses. Save them and regenerate.

Recap: Only trusted CI scans count · Pick projects and an audit period; charged once · Controls tested as x of y, in SOC 2 wording · It says what it couldn't test · Send the external summary; auditors can verify it Next: Chapter 23 · AI review

Review AI findings and app safeguards

Use optional model review and inspect safeguards in your AI app.

Add model review verdicts and inspect their reasons.

Current guide: AI review uses your configured model provider. Check what information will be shared before enabling it.

Read the transcript

AI · Chapter 23: AI review Let an AI double-check what Skylos found. Nothing is hidden, and blocking never changes.

Step 1: A second opinion on every issue Skylos finds the issues. Your model reviews Python security issues and unused code in every language, and each one gets a verdict.

Step 2: Add --ai-review to a scan The scan runs as usual, then your model reviews the issues. One line sums up what it thought.

Step 3: Every verdict has a reason In JSON, each issue carries the AI's verdict, the model, how many runs it took and why.

Step 4: Send it to Skylos Cloud skylos cloud --ai-review uploads the scan with the verdicts. The SQL injection still fails the gate.

Step 5: See the verdict on the scan Each issue the AI reviewed carries its verdict, right in the issue list.

Step 6: Open an issue for the reason The AI review panel shows which model looked, how many times, and why it agrees or disagrees.

Step 7: Turn it on in CI Default-branch scans from CI set each issue's AI state in Issues and on the Pull requests board. Pull request jobs never get the key.

Recap: --ai-review adds a verdict to each issue · Nothing is hidden and blocking never changes · Every verdict says which model and why · skylos cloud --ai-review shows it on the scan · Turn it on in CI with one variable and a key Next: Chapter 24 · AI app safety

Inspect detected model calls and review missing safeguard patterns.

Current guide: Checks and scores describe the patterns assessed. They do not prove that an AI app is safe.

Read the transcript

AI · Chapter 24: AI app safety Check the AI features in your own app for missing safeguards before they ship.

Step 1: Your app calls an LLM. Is it safe? Skylos finds every LLM call and agent tool in your code, checks each one for safeguards, and scores it.

Step 2: Find the AI features in your code The new support assistant calls OpenAI. Skylos shows where, what feeds it and whether it can reach anything dangerous.

Step 3: Check its safeguards Each check passes or fails with a reason. Raw output is used without validation and the model name can change under you.

Step 4: Send it to Skylos Cloud skylos defend . --upload records the result, so the whole team sees it.

Step 5: Every repository's AI app safety in one place AI, then AI app safety: each repository's latest score, how many checks pass and how many AI features it has.

Step 6: See what's missing and how to add it Open a result: missing safeguards come first, each with the line and how to add it. What already passes is on the right.

Step 7: Keep it current in CI skylos cicd init --defend adds the check to CI, so every push to the default branch updates this page.

Recap: skylos discover finds every LLM call and agent tool · skylos defend checks each one for safeguards · Missing safeguards come with how to add them · AI app safety shows every repository's score · Add it to CI with skylos cicd init --defend Next: Guides at skylos.dev/guides

Plans and credits

Use the current pricing guide

Find the current workspace plans, team allowances and credit costs before choosing a plan or using a paid feature.

Read plans and pricing