Chat Rule Sets

A chat rule set is the root-level sibling of webhook and schedule rule sets. Instead of reacting to an inbound event, it lists the marketplace plugins the chat tool (SupervisorSubagentTools.RunClaudeCodeOnRepository) may offer the user, together with the cost/control tuning to apply to those chat dispatches.

It is discriminated by a non-empty chat key, so the same rules.json array can hold webhook, schedule, and chat rule sets side by side — the webhook evaluator keeps only "webhook" entries, so chat rule sets never leak into it.

{
  "chat": "chat",
  "model": "claude-sonnet-4-5",
  "max-budget-usd": 5.0,
  "with-envs": [
    { "name": "GITHUB-TOKEN", "value": "secrets.GITHUB-TOKEN", "mandatory": true }
  ],
  "use-plugins": [
    {
      "plugin-name": "pr-reviewer@xianix-plugins-official",
      "marketplace": "xianix-team/plugins-official",
      "slash-command": "/pr-review"
    }
  ]
}

No `executions` array

Unlike a webhook or schedule rule set, a chat rule set has no executions. A chat run’s prompt is authored by the supervisor as {slash-command} {user-target} from the slash-command on each listed plugin — there’s nothing to interpolate from a payload, so a per-execution execute-prompt / use-inputs / match-any shape would be redundant.

The tuning knobs and plugin list therefore live at the rule-set root and apply uniformly to every chat dispatch that uses one of the listed plugins. Author separate chat rule sets when different plugins need different tuning.

`chat` — discriminator & label

"chat": "chat"

Any non-empty value marks the object as a chat rule set. The value itself is only used for logs and diagnostics — there is no external event name to match (unlike a webhook rule set’s webhook).

`use-plugins` with `slash-command`

use-plugins uses the same schema as webhook/schedule rule sets, but here the slash-command field is important: it’s how the supervisor knows the exact command to compose for the chat dispatch. Chat plugins should always declare one so the supervisor does not have to invent a command name.

"use-plugins": [
  {
    "plugin-name": "pr-reviewer@xianix-plugins-official",
    "marketplace": "xianix-team/plugins-official",
    "slash-command": "/pr-review"
  }
]

Field	Required (chat)	Description
`plugin-name`	Yes	Plugin reference in `plugin-name@marketplace-name` form.
`marketplace`	No	Marketplace source. Omit for the built-in Anthropic marketplace.
`slash-command`	Recommended	The Claude Code slash command the supervisor invokes (e.g. `/pr-review`). A chat listing’s `slash-command` overrides one omitted on a webhook listing of the same plugin.

Rule-set-level tuning

The Cost & Execution Controls sit at the chat rule-set root (not inside an execution block) and apply to every chat dispatch that picks one of this set’s plugins:

Field	Description
`model`	Model override for chat dispatches. Empty means the executor’s default.
`max-turns`	Turn cap for chat dispatches.
`allowed-tools`	Allow-list of tools for chat dispatches.
`disallowed-tools`	Deny-list of tools for chat dispatches.
`max-budget-usd`	Per-run USD budget cap for chat dispatches.
`resume-sessions`	Whether chat dispatches resume prior sessions.

Rule-set-level `with-envs`

with-envs at the chat rule-set root is applied to every chat dispatch, mirroring the webhook / schedule rule-set commons. It’s platform-agnostic — the same three value forms (secrets.*, host.*, constant) apply.

"with-envs": [
  { "name": "GITHUB-TOKEN", "value": "secrets.GITHUB-TOKEN", "mandatory": true }
]

How chat plugins are surfaced

Chat rule sets feed two supervisor tools via a deduplicated plugin catalog:

ListAvailablePlugins — lets the chat model discover which plugins exist, the slash command each exposes, and which environment variables the tenant must have configured.
RunClaudeCodeOnRepository — resolves a chosen plugin name back to its full spec (and the with-envs declared alongside it), validates that mandatory inputs are supplied, then dispatches the run. This is the chat-side equivalent of the input validation the webhook evaluator performs.

A few consequences worth knowing:

Chat-exclusive, else webhook fallback. If a plugin is listed by any chat rule set, the catalog serves it exclusively through the chat rule set’s tuning. A plugin only ever referenced by webhook rule sets falls back to those webhook usage examples (unchanged behaviour).
Structural inputs are auto-filled. repository-url and repository-name are resolved by the chat tool from the repository the user chose — the model never supplies them. repository-name is derived from the clone URL via the same platform-aware logic the webhook path uses.
No prompt to interpolate. The chat usage example carries only the tuning knobs (no execute-prompt, no inputs) — the supervisor authors the prompt from the user’s message.

Complete Example

A chat rule set exposing the PR reviewer to chat with its own tuning, alongside a webhook rule set in the same rules.json array:

[
  {
    "webhook": "Default",
    "with-envs": [
      { "name": "GITHUB-TOKEN", "value": "secrets.GITHUB-TOKEN", "mandatory": true }
    ],
    "executions": [
      {
        "name": "github-pull-request-review",
        "platform": "github",
        "repository": "repository.clone_url",
        "match-any": [
          { "name": "pr-opened", "rule": "action==opened" }
        ],
        "use-plugins": [
          {
            "plugin-name": "pr-reviewer@xianix-plugins-official",
            "marketplace": "xianix-team/plugins-official"
          }
        ],
        "execute-prompt": "You are reviewing pull request #{{pr-number}} in {{repository-name}}.\n\nRun /pr-review {{pr-number}} to perform the automated review."
      }
    ]
  },
  {
    "chat": "chat",
    "model": "claude-sonnet-4-5",
    "max-budget-usd": 5.0,
    "use-plugins": [
      {
        "plugin-name": "pr-reviewer@xianix-plugins-official",
        "marketplace": "xianix-team/plugins-official",
        "slash-command": "/pr-review"
      }
    ]
  }
]

Here the pr-reviewer plugin is available both automatically (via the webhook rule set on action==opened) and interactively (via the chat rule set). Because it’s listed under a chat rule set, chat dispatches use the chat rule set’s Sonnet model and $5.00 budget cap rather than falling back to the webhook execution’s tuning.

Chat Rule Sets

No executions array

chat — discriminator & label

use-plugins with slash-command