agentleFS
AlexisBalayreAlexisBalayre/claude-code-config/.claude/skills/wiz/SKILL.md

wiz

Skill2 starsChanged 29 days agoMITView on GitHub

Investigate the project's cloud security posture through the Wiz MCP (read-only), routing each query to the right domain and scoping to the project's own resources.

  • Reads credentials

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

6 rules. About 0 are common, 1 is distinctive, and 5 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 5

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Every query takes the same four steps:

    line 20
  2. Very specific:

    Route: Call discover() with no argument to list the domains.

    line 22
  3. Very specific:

    Spec: Call discover(domain) for the exact tool name and its parameter schema.

    line 23
  4. Very specific:

    Query: Call execute(tool_name, parameters) with the exact name from step 2, scoped per the project section below.

    line 24
  5. Very specific:

    Prefer a *_grouped tool with a small first over a raw list: fewer tokens, and it answers "where is the load" directly.

    line 27
  6. Very specific:

    Default severity to ["CRITICAL", "HIGH"], status to ["OPEN"], and add has_fix: true when the goal is to fix (only fixable findings are …

    line 52

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Wiz (cloud security)

The Wiz MCP exposes ~250 tools across 24 security domains behind two gateway tools. You never call a Wiz tool directly: discover finds it, execute runs it. This skill is read-only: investigate, never mutate.

Setup: this skill needs the Wiz MCP server enabled in .claude/settings.local.json and your own Wiz resource IDs in .env (see .env.example): WIZ_REPO_BRANCH_ID (the default-branch REPOSITORY_BRANCH resource for SAST) and WIZ_CONTAINER_REPO_IDS (the container-repository UUIDs for your images). Resolve them once per the "Resolve to IDs" section below, then fill them in — never hardcode real UUIDs in the repo.

The loop

Every query takes the same four steps:

  1. Route. Call discover() with no argument to list the domains. Pick the one domain that owns the answer by the type of finding you want, not by nouns in the question (see Routing traps).
  2. Spec. Call discover(domain) for the exact tool name and its parameter schema. Read it: params that sound singular are usually arrays (group_by, severity, status, asset_name, container_repository), some are required (e.g. list_issues_grouped needs group_by + type), and the first / limit cap differs per tool (often 20). Never guess.
  3. Query. Call execute(tool_name, parameters) with the exact name from step 2, scoped per the project section below. If execute rejects the name, re-run discover(domain) to confirm spelling (some names repeat across domains).
  4. Summarize. A result too large for one response is auto-saved to a tool-results/*.txt file instead of returned. Parse it with jq or python (or hand the file to a subagent), never re-read it whole. Report a ranked summary (severity, then affected resource) with the Wiz finding IDs so the user can open each in the portal. Never paste raw JSON back.

Prefer a *_grouped tool with a small first over a raw list: fewer tokens, and it answers "where is the load" directly.

Routing traps

Route by finding type. discover()'s own text carries "Do NOT use for X → Y" notes; these are the most-often mis-routed:

  • "AI" in the question does not mean the ai_security domain. Route by what you want about AI: issues to issues, threats to soc, SAST to codesec, CVEs to vulnerabilities, sensitive data to dspm, compliance to compliance. Use ai_security only for the AI resource inventory itself.
  • Third-party CVE on a cloud resource or image goes to vulnerabilities; a first-party source-code weakness (SAST) goes to codesec. "Issues in our images / repo" spans both: image CVEs are vulnerabilities, code flaws are codesec. Check both when the ask is broad.
  • A compliance framework score or cross-framework gap goes to compliance; a single misconfiguration or security rule goes to cspm.
  • Cloud IAM (principals, entitlements, access keys, MFA) goes to identity; Wiz portal users and RBAC go to settings.
  • EOL, outdated, or unsupported technology goes to inventory, not container_security.

Resolve to IDs (names lie)

A free-text name filter that returns zero does not mean the resource is absent: asset_name: ["<project name>"] can return nothing even though the project's images exist. Filter by the resource's Wiz UUID instead.

The fastest UUID source is a Wiz portal URL: its hash fragment encodes the exact filter IDs the view uses (resource, containerRepository, ...). Ask the user to paste the portal link for the view they mean, or read the IDs out of one they already sent. Otherwise resolve the UUID with a listing / inventory tool, then filter by it.

The project in Wiz

Typically the project's code and images live under the source host and container registry (e.g. GitHub / ghcr.io), not the cloud account; only the running workloads are on the cloud provider (cloud_platform: ["AWS"] or equivalent, plus your prod account, cluster, and namespace). So a cloud-platform filter hides the image and SAST findings: scope those by the IDs below, not by cloud platform.

  • Source repo (SAST → codesec / list_sast_findings): resource_id: ["${WIZ_REPO_BRANCH_ID}"] (the default-branch REPOSITORY_BRANCH resource). Add is_default_branch: true.
  • Container images (CVEs → vulnerabilities): filter container_repository by the UUIDs in ${WIZ_CONTAINER_REPO_IDS} — one per image the project ships.

Default severity to ["CRITICAL", "HIGH"], status to ["OPEN"], and add has_fix: true when the goal is to fix (only fixable findings are actionable). Per finding, layerMetadata.isBaseLayer and artifactType.group (OS_PACKAGE vs CODE_LIBRARY) say whether the fix is a base-image bump or a dependency bump.

execute("list_vulnerability_findings_grouped", {
  group_by: ["VULNERABLE_ASSET"], asset_type: ["CONTAINER_IMAGE"],
  container_repository: ["<UUID from $WIZ_CONTAINER_REPO_IDS>"],
  severity: ["CRITICAL", "HIGH"], status: ["OPEN"], has_fix: true, first: 20 })

Before treating a finding as real

Wiz findings, SAST especially, carry false positives; verify against the code before proposing a fix. Read the flagged line. Example false-positive shapes: a weak hash (SHA-1, MD5) mandated by a protocol or used as a non-security checksum, a timer or callback API flagged as "eval", internal config dirs flagged as path traversal, and template rendering with escaping off that produces plain text rather than HTML.

Read-only guardrail

This skill only investigates. Execute only read tools: names beginning list_, get_, or search_, plus discover.

Refuse any mutating tool: anything that creates, changes, resolves, assigns, deletes, reports, or configures (creating a pentest finding, an automation or remediation rule, a Jira or ServiceNow ticket, changing an issue's status). If the user wants one, say it is out of scope for this skill and point them to the Wiz portal.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.