Skip to main content
The AI does not know your runbooks, your conventions, or your internal APIs. Capabilities is where you teach it: skills are reusable markdown instructions it follows, and custom tools are sandboxed scripts it can call. Both live under AI → Settings → Capabilities.
This page covers what Ankra’s own AI uses inside the platform. To teach your editor or assistant (Claude Code, Cursor, Copilot and others) Ankra’s conventions, install the Agent Skills instead.

​
Skills

Skills are reusable markdown instructions the AI draws on during conversations - runbooks, add-on configuration templates, CI/CD pipeline conventions. Ankra ships a set of platform skills, and organisation admins can add their own or override the built-ins.

​
How skills resolve

Skills come in two tiers:
  • Platform skills - built-in, read-only, maintained by Ankra.
  • Organisation skills - created by your admins, scoped to your organisation.
An organisation skill with the same name as a platform skill shadows it - the AI uses your version instead. The skills page shows an Override badge on shadowing skills and a line-level diff against the platform original, with a one-click Discard override to restore the built-in.

​
Creating a skill

1

Open the editor

Go to AI → Settings → Capabilities and click New skill - or select a platform skill and click Override to start from its content.
2

Fill in the fields

  • Name - lowercase letters, digits, hyphens/underscores (max 100 chars); matching a platform skill’s name overrides it
  • Description - what the skill teaches (max 500 chars)
  • Content - the markdown instructions (max 200,000 chars); YAML frontmatter at the top marks add-on and CI/CD skills
  • Kind - core, cicd, or addon
  • Add-on charts - for addon skills, the chart names the skill applies to
  • Always active - inject the skill into every conversation instead of on demand, and into every Org skills review on repositories where that lane is on
3

Save

The skill takes effect immediately - the AI’s skill cache is refreshed on every change.

​
Who can change skills

All members can browse skills; creating, editing, and deleting requires organisation admin. Platform skills cannot be edited or deleted - only overridden.

​
Custom tools

Custom tools extend what the AI can do with capabilities specific to your organisation: a security scan, an internal API call, a report generator. Each tool is a script with a typed input schema; once enabled, the AI can call it like any built-in tool.

​
The sandbox model

Tool runs execute as one-shot sandbox jobs on your staging cluster, isolated from the rest of the platform:
  • Runs use a signed runner image with no Kubernetes API access.
  • Network egress is limited to DNS and HTTPS.
  • CPU, memory, and execution time are capped.
  • The AI’s arguments arrive as JSON in the TOOL_INPUT environment variable, validated against your input schema first.

​
Creating a tool

1

Define the interface

Name the tool (lowercase, digits, underscores), describe what it does and when the AI should use it, and provide a JSON Schema for its input. The description is what the AI reads when deciding to call it - make it concrete.
2

Write the script

Pick a runner and write the script. Read TOOL_INPUT for arguments; whatever the script prints is returned to the AI.
3

Set the guardrails

Choose a risk level (low, medium, high), optionally mark the tool trusted, and scope it to all clusters or named clusters only. Secret values come from AI chat secret slots mounted as files - never hardcode credentials in the script.
Every save creates a new version; the version history on each tool shows what changed and when. Tools can be disabled without deleting them - a disabled tool is not offered to the AI.

​
Who can change tools

Creating, editing, enabling, and deleting custom tools requires organisation admin. All members can see the tool list and version history.

​
Next

Per-component learnings that travel with one add-on or manifest belong in AGENTS.md, not in a skill.