llm-orchestrator 1.3.1 → 1.5.0

package/schemas/agent-roles.schema.json

@@ -54,6 +54,7 @@
           "skills",
           "mcps",
           "best_for",
+          "charter",
           "permission_profile",
           "default_tier"
         ],
@@ -79,13 +80,99 @@
           },
           "mcps": {
             "type": "array",
+            "description": "Real MCP server ids only (an id or alias of an installable MCP in registries/preferred-tools.json). A capability whose implementation the project binds goes in tool_capabilities instead.",
             "items": {
-              "type": "string"
+              "type": "string",
+              "pattern": "^[a-z0-9][a-z0-9_-]*$"
+            }
+          },
+          "tool_capabilities": {
+            "type": "array",
+            "description": "Capability classes the role needs whose concrete tool the project binds (a database client, a provider API). Rendered as capabilities, never as a server name to search for.",
+            "items": {
+              "type": "object",
+              "required": [
+                "capability",
+                "label"
+              ],
+              "properties": {
+                "capability": {
+                  "type": "string",
+                  "pattern": "^[a-z][a-z_]*\\.[a-z][a-z_.]*$",
+                  "description": "Capability id, as used in registries/preferred-tools.json and registries/task-mappings.json."
+                },
+                "label": {
+                  "type": "string",
+                  "minLength": 1,
+                  "pattern": "^[^\\r\\n`]+$",
+                  "description": "Plain-language name of the capability class, e.g. 'database client'."
+                }
+              },
+              "additionalProperties": false
             }
           },
           "best_for": {
             "type": "string"
           },
+          "charter": {
+            "type": "object",
+            "description": "Role-specific operating charter rendered into the body of each native agent file by adapters/agents.mjs. Every string is one line of plain prose.",
+            "required": [
+              "mission",
+              "principles",
+              "done_when",
+              "never",
+              "handoff"
+            ],
+            "properties": {
+              "mission": {
+                "type": "string",
+                "minLength": 1,
+                "pattern": "^[^\\r\\n]+$",
+                "description": "One-sentence identity, in second person."
+              },
+              "principles": {
+                "type": "array",
+                "minItems": 3,
+                "maxItems": 5,
+                "items": {
+                  "type": "string",
+                  "minLength": 1,
+                  "pattern": "^[^\\r\\n]+$"
+                },
+                "description": "Concrete operating rules specific to this role."
+              },
+              "done_when": {
+                "type": "array",
+                "minItems": 2,
+                "maxItems": 4,
+                "items": {
+                  "type": "string",
+                  "minLength": 1,
+                  "pattern": "^[^\\r\\n]+$"
+                },
+                "description": "Observable exit criteria."
+              },
+              "never": {
+                "type": "array",
+                "minItems": 2,
+                "maxItems": 4,
+                "items": {
+                  "type": "string",
+                  "minLength": 1,
+                  "pattern": "^[^\\r\\n]+$"
+                },
+                "description": "Role-specific prohibitions."
+              },
+              "handoff": {
+                "type": "string",
+                "minLength": 1,
+                "pattern": "^[^\\r\\n]+$",
+                "description": "What the role returns to its parent."
+              }
+            },
+            "additionalProperties": false
+          },
           "permission_profile": {
             "type": "string"
           },

package/schemas/capability-contract.schema.json

@@ -25,6 +25,7 @@
       "required": ["type"],
       "properties": {
         "type": { "type": "string", "minLength": 1 },
+        "domain": { "type": "string", "description": "The pre-evaluation domain (FRONTEND, BACKEND, DB, INFRA, UNKNOWN or a project domain); matched case-insensitively by any_task_domains." },
         "signals": { "type": "array", "items": { "type": "string" }, "uniqueItems": true },
         "requires_shell": { "type": "boolean" },
         "nontrivial": { "type": "boolean" },

package/workflows/bug-fix.md

@@ -11,6 +11,10 @@
   reproduction, does not pass G1.
 - Exactly one falsifiable hypothesis at G2, recorded in `hypothesis:{task_id}`.
 - A regression test that is **observed RED** before the fix, then GREEN after.
+- **`impeccable` when the defect is in UI or visual design** (`ui.design_system`): **Evidence** —
+  `/impeccable audit` on the affected surface; **Fix** — `/impeccable harden` for error, overflow,
+  i18n and edge-case defects; **Verify** — `/impeccable audit` again on the same surface. Another
+  design skill never stands in for it silently.
 
 ## Optional (trigger)
 

package/workflows/feature.md

@@ -11,11 +11,19 @@
 - **The test-engineer writes the failing test first.** No builder is dispatched before G3 RED.
 - Discover project conventions and acceptance criteria before designing against them.
 - Independent review wherever a risk floor applies.
+- **`impeccable` on any UI** (domain `FRONTEND`, or the feature creates or changes UI or visual
+  design — `ui.design_system`): **Plan** — `/impeccable init` when the project has no `PRODUCT.md`,
+  then `/impeccable shape` before any UI code; **Build** — an ordinary `/impeccable` new-work
+  request after `shape`, or the targeted command the plan names (`layout`, `typeset`, `colorize`,
+  `animate`, `adapt`, …); **Review** — `/impeccable critique` + `/impeccable audit` on the changed
+  surface; **Verify** — `/impeccable polish` before G5. Design-system, UI-styling,
+  `frontend-design` and `ui-ux-pro-max` skills are complements, never a substitute.
 
 ## Optional (trigger)
 
 - `context7` — any library or SDK API surface, **including familiar ones**; training data lags.
-- Design-system + UI-styling skills — any new UI.
+- Design-system + UI-styling skills — alongside `impeccable` when the project has its own design
+  system to follow.
 - Payment-provider best-practices skill — the change touches a money path.
 - Semantic-code MCP — a large cross-file edit.
 

package/workflows/investigation.md

@@ -19,6 +19,10 @@
   explicit refusal continue in declared degraded mode with a `degraded: exa` line and those claims
   labelled unverified.
 
+- **`impeccable` when the question is about UI or visual design** (`ui.design_system`): **Evidence**
+  — `/impeccable critique` and `/impeccable audit` on the surface in question, read-only; the
+  report cites their findings.
+
 ## Optional (trigger)
 
 - Semantic-code MCP — symbol-level questions across a large codebase.

package/workflows/refactor.md

@@ -11,6 +11,10 @@
 - Identify behavior and compatibility invariants before touching anything. Externally visible
   contracts are preserved — no renamed error keys, no removed response fields, no moved routes.
 - Small transformations with disjoint ownership, each verifiable on its own.
+- **`impeccable` when the refactor touches UI** (`ui.design_system`): **Analyze** —
+  `/impeccable audit` for the baseline; **Incremental** — `/impeccable extract` to pull repeated styles into
+  tokens and components; **Review** — `/impeccable audit` against the baseline (no new findings)
+  and `/impeccable polish` for design-system alignment. Never a silent substitute.
 
 ## Optional (trigger)
 

package/workflows/review.md

@@ -15,6 +15,10 @@
   described as independent.
 - Minor findings and out-of-scope calls are batched into one explicit "skip or fix?" question — never
   silently parked.
+- **`impeccable` when the diff or target is UI or visual design** (`ui.design_system`): **Analyze** —
+  `/impeccable critique` (UX review) + `/impeccable audit` (accessibility, performance, responsive,
+  anti-pattern detectors); their findings join the severity-ranked list. A guidelines skill
+  complements them, never replaces them.
 
 ## Optional (trigger)