Skip to main content

Skills

/skills is the capability catalogue: everything agents can do, from built-in tools to custom skill packages, managed in one place. Sidebar items are role-filtered — Skills requires the skills curator role or above. For authoring and packaging skills, see Custom skills.

The page has five tabs — System Skills, Marketplace Skills, CLI Skills, Custom Skills, and Settings — plus a search box that filters the active tab (the search is separator-insensitive, so "list squad" matches list_squad).

System Skills​

Org-wide management of the built-in tools every agent can use.

  • Tool cards — each card shows the tool name, category badge, description, effective timeout (flagged as an override when an admin has changed it from the default), an optional admin note, and a Requires Approval badge where applicable. A Configured badge marks tools with saved settings, and a "Linked custom skills" line lists any custom skills that auto-mount into that tool.
  • Enable/disable — a per-card toggle; disabling a tool removes it from all agents org-wide. Enable all / Disable all buttons bulk-toggle, and when a search or category filter is active they scope to only the visible tools, with a confirmation dialog spelling out exactly what will change.
  • Configure — expands a per-tool panel with an admin note and the tool's configuration fields: API keys, default limits, blocked domains or commands, timeouts, and other guardrails. Settings saved here apply as org-wide defaults for every agent.
  • Filters — category chips, a Show disabled toggle, and an enabled/total counter.

Marketplace Skills​

The approved skill catalogue for the organisation.

  • Catalogue rows — approved skills that can be installed onto agents. Install opens a picker of active agents and installs the skill onto the one you choose.
  • Marketplace packages — upload a marketplace package (.zip) by drag-and-drop or file picker. Uploads are staged first (they survive reloads), then a per-row Install applies the package's components; a Test action verifies it, and a SPEC viewer shows the package's specification document.
  • Browse marketplace — an in-app drawer for browsing the configured remote marketplace without leaving the page.
  • Deep links from other pages land on this tab and highlight the referenced skill row.

CLI Skills​

Command-line capabilities that run in sandboxed containers.

  • Native CLI skills — the built-in set (shell, PowerShell, AWS/Azure/Google Cloud CLIs, kubectl, Terraform, Docker, SQL, Git, curl, Python, Node.js, and the Claude Code and Codex CLI agents), each listed with its container image and a risk level from low to critical.
  • Create a CLI skill — teach agents a new CLI by submitting its name, the CLI tool, an optional version, and source material (such as its help text). The platform runs a safety analysis that surfaces destructive, credential-related, and network commands and assigns a risk level.
  • Row actions — view the generated skill definition and safety analysis, Publish a draft so agents can use it, and Delete with confirmation.

Custom Skills​

Your organisation's own skill packages and their review pipeline.

  • Upload — drag-and-drop or pick a Claude-format skill package (.zip or .skill). Uploaded skills land in the consolidated skills table as private skills.
  • Review pipeline — submitted skills progress through Submit → SBOM Generation → Static Analysis → CVE Scan → Build Verify → Sandbox Test → Review → Approved, with the pipeline stage visible per skill. Reviewers Approve (optional comment) or Reject (reason required).
  • Status filters — All, Submitted, Reviewing, Approved, Deprecated, Revoked.
  • Verify mount — a deterministic check that resolves and materialises every skill server-side and returns the actual file list and sandbox paths, so you can trust a skill will mount without running an agent.
  • Auto-apply — a per-skill toggle controlling whether the skill auto-mounts into the tools it applies to (read-only for marketplace-installed skills).
  • Delete — a confirmation modal that, for published skills, also explains the cascade: the marketplace listing is deprecated and active installs elsewhere are uninstalled, with a summary shown after deletion.

Settings​

Sandbox execution configuration for skill and CLI runs.

  • Docker status — whether the sandbox engine is connected, the default container image, and a Test Sandbox button that runs a live check and reports success and duration.
  • Resource and security policy — per-risk-level resource limits, plus toggles for which system tools and language runtimes are enabled in the sandbox.
  • Save persists the configuration org-wide.