Live-skills (`/skills` slash command)

The Skills Gateway page (UI → Skills → Skills Gateway) renders a copy-pasteable slash command that lets a coding agent (Claude Code, Cursor, Spec Kit, etc.) browse, search, run, install, and update skills served by the CAIPE skill catalog.

The body of that slash command is rendered from a single Markdown template called the live-skills skill. This page describes how operators can customize it for their deployment without forking the chart or rebuilding the image.

Where the template lives

Layer	Location
Packaged default	`charts/ai-platform-engineering/data/skills/live-skills.md`
Helm ConfigMap	`skills-live-skills` (key `live-skills.md`)
Mounted in pod	`/app/data/skills-live-skills/live-skills.md` on the `caipe-ui` pod
Served at	`GET /api/skills/live-skills` (returns `{ template, source, … }`)
Rendered in	`Skills Gateway` UI → "Create the live-skills skill" card

The UI substitutes three placeholders client-side based on the form fields on the page:

Placeholder	Replaced with
`{{COMMAND_NAME}}`	Slash command name (default `skills`)
`{{DESCRIPTION}}`	Description shown in the slash-command picker
`{{BASE_URL}}`	Gateway base URL (auto-detected from the request origin)
`{{ARG_REF}}`	Per-agent argument syntax (`$ARGUMENTS`, `$1`, or `{{input}}`)

{{ARG_REF}} is substituted per agent by the renderer (see Multi-agent support below) so the same canonical template produces the right $ARGUMENTS (Claude/Cursor/Spec Kit), $1 (Codex/Gemini), or {{input}} (Continue) reference for each surface.

Override resolution order

The UI server (/api/skills/live-skills) picks the first source that resolves:

SKILLS_LIVE_SKILLS_TEMPLATE env var (raw markdown) — highest priority
File at SKILLS_LIVE_SKILLS_FILE env var (default: /app/data/skills-live-skills/live-skills.md, mounted by the chart)
Packaged default at data/skills/live-skills.md (chart-relative, dev only)
A minimal built-in fallback string

The currently active source is shown in the UI under "Template source: …" so operators can verify which override is in effect.

Override option 1 — inline Helm value (simplest)

Drop a multi-line string into your Helm values:

skillsLiveSkills: |
  ---
  description: My company's skill catalog
  ---

  ## User Input

  ```text
  $ARGUMENTS
  ```

  ## SECURITY — never expose the API key
  - NEVER print, echo, or display the API key in any output.

  ## Steps

  1. Fetch from {{BASE_URL}}/api/skills with the X-Caipe-Catalog-Key header.
  2. Render results as a table.
  3. Use `/{{COMMAND_NAME}} <query>` to search, `/{{COMMAND_NAME}} run <name>`
     to execute inline, or `/{{COMMAND_NAME}} install <name>` to save locally.

The chart writes that string verbatim into the skills-live-skills ConfigMap. No image rebuild required.

Override option 2 — named variant shipped with the chart

If you want to ship a small set of opinionated variants with your fork of the chart (e.g. one per business unit):

Drop additional files alongside live-skills.md, named live-skills.<variant>.md — for example live-skills.enterprise.md.
Select the variant via Helm:
```
skillsLiveSkillsName: enterprise
```

The template resolves data/skills/live-skills.<variant>.md and falls back to live-skills.md if that file isn't found.

Override option 3 — bring-your-own ConfigMap or env

For deployments that consume the published chart unchanged:

Mount your own ConfigMap

caipe-ui:
  volumes:
    - name: my-live-skills
      configMap:
        name: my-custom-skills-live-skills
  volumeMounts:
    - name: my-live-skills
      mountPath: /app/data/skills-live-skills
      readOnly: true

The default SKILLS_LIVE_SKILLS_FILE (already set by the chart) keeps pointing at /app/data/skills-live-skills/live-skills.md, so the new ConfigMap content takes effect immediately.

Or stuff the markdown into an env var

caipe-ui:
  env:
    SKILLS_LIVE_SKILLS_TEMPLATE: |
      ---
      description: Inline override via env
      ---
      ## Steps
      1. Fetch from {{BASE_URL}}/api/skills.

This wins over any file-based source and is handy for quick experiments.

Or point at a different file

caipe-ui:
  env:
    SKILLS_LIVE_SKILLS_FILE: "/etc/caipe/my-live-skills.md"

Useful when mounting a Secret or a sidecar-managed file at a non-default path.

What the user sees

Once the template is in place, the Skills Gateway page lets the user:

Pick a slash command name (skills by default).
Pick a description (rendered into the artifact's frontmatter / metadata).
Pick a coding agent (Claude Code, Cursor, Spec Kit, Codex CLI, Gemini CLI, Continue) — the install path, file format, and argument syntax are derived from this choice.
Copy the generated install command (mkdir -p … && cat > … << 'SKILL' …) or, for Continue, the JSON fragment to merge into ~/.continue/config.json.
Preview the rendered artifact (Markdown / TOML / JSON).
Read the per-agent launch & invocation guide rendered just below the install command.

The canonical template is rendered server-side per agent, so a single ConfigMap serves every surface without operators maintaining N copies.

Multi-agent support

GET /api/skills/live-skills?agent=<id>&command_name=<name>&description=<desc> returns a per-agent rendered artifact plus install/launch metadata. The agent registry currently ships with six entries:

Agent ID	Label	Install path	Format	Argument syntax
`claude`	Claude Code	`.claude/commands/{name}.md`	Markdown + frontmatter	`$ARGUMENTS`
`cursor`	Cursor	`.cursor/commands/{name}.md`	Markdown + frontmatter	`$ARGUMENTS`
`specify`	Spec Kit	`.specify/templates/commands/{name}.md`	Markdown + frontmatter	`$ARGUMENTS`
`codex`	Codex CLI (OpenAI)	`~/.codex/prompts/{name}.md`	Plain Markdown	`$1`
`gemini`	Gemini CLI	`~/.gemini/commands/{name}.toml`	TOML (`description`, `prompt`)	`$1`
`continue`	Continue (VS Code / JetBrains)	`~/.continue/config.json` (fragment to merge)	JSON fragment	`{{input}}`

The renderer parses the canonical Markdown's frontmatter once, substitutes {{COMMAND_NAME}}, {{DESCRIPTION}}, {{BASE_URL}}, and {{ARG_REF}}, then re-wraps the body as appropriate for each surface (YAML frontmatter, TOML basic strings, or a JSON object). Adding a new agent is one entry in ui/src/app/api/skills/live-skills/agents.ts plus a case in renderForAgent().

Per-agent launch & invocation guidance

The UI renders a short "Launch <Agent> and use it" panel after the install command, summarizing how to install and invoke the chosen agent. The exact text is part of each agent's spec (launchGuide) and supports basic Markdown (bold, inline code, links, fenced code blocks).

Quick reference

Claude Code — npm install -g @anthropic-ai/claude-code → claude from your repo root → /skills. Auto-discovers commands in .claude/commands/ (per-repo) and ~/.claude/commands/ (user-global).
Cursor — install from cursor.com, open the repo, then Cmd/Ctrl + L → /skills. Reload the window if a new command doesn't appear in the picker.
Spec Kit — uvx --from git+https://github.com/github/spec-kit.git specify init. Spec Kit re-syncs commands into the agent-specific directory the next time you run /specify, /plan, /tasks, or /implement.
Codex CLI — npm install -g @openai/codex → codex → /skills. Prompts live in ~/.codex/prompts/ (user-global). Use $1-style positional args when invoking the prompt.
Gemini CLI — npm install -g @google/gemini-cli → gemini from your repo root → /skills "kubernetes" (quote multi-word args). Commands live in ~/.gemini/commands/ (user-global) or .gemini/commands/ (per-repo).
Continue — install the VS Code or JetBrains extension, then merge the rendered JSON fragment into the top-level slashCommands array of ~/.continue/config.json. Continue reloads config.json automatically.

Where the template lives​

Override resolution order​

Override option 1 — inline Helm value (simplest)​

Override option 2 — named variant shipped with the chart​

Override option 3 — bring-your-own ConfigMap or env​

Mount your own ConfigMap​

Or stuff the markdown into an env var​

Or point at a different file​

What the user sees​

Multi-agent support​

Per-agent launch & invocation guidance​

Quick reference​

See also​