Add per-role effort configuration to Quest

.ai/allowlist.json

@@ -23,6 +23,17 @@
   "codex_fallback_model": "gpt-6-astra",
   "codex_auth_mode": "cached",
   "codex_reasoning_effort": "medium",
+  "effort": {
+    "planner": "medium",
+    "plan-reviewer-a": "medium",
+    "plan-reviewer-b": "medium",
+    "arbiter": "medium",
+    "builder": "medium",
+    "code-reviewer-a": "medium",
+    "code-reviewer-b": "medium",
+    "review-arbiter": "medium",
+    "fixer": "medium"
+  },
   "models": {
     "planner": "gpt-6-astra",
     "plan-reviewer-a": "claude-opus-5",

.ai/quest.md

@@ -38,7 +38,7 @@ The Quest Agent interprets your intent, matches brief references, and routes to
 
 ## Roles
 
-Role models come from `.ai/allowlist.json` at startup and the saved `.quest/<id>/orchestration.json` thereafter. The canonical dispatch matrix in `.skills/quest/delegation/workflow.md` selects the runtime and transport; role names do not imply a fixed model.
+Role models and per-role `effort` come from `.ai/allowlist.json` at startup and the saved `.quest/<id>/orchestration.json` thereafter. Active Claude/Codex roles require an effort pin; Gemini roles omit it. Old saved quests without the map, or with the retired saved `codex_reasoning_effort` scalar, stop before dispatch: explicitly reconfigure saved settings or start a new quest. The allowlist scalar remains a standalone `/gpt` default only. The canonical dispatch matrix in `.skills/quest/delegation/workflow.md` selects the runtime and transport; role names do not imply a fixed model.
 
 | Role | Instructions | Purpose |
 |------|--------------|---------|
@@ -96,7 +96,7 @@ Codex-led Quest note:
   | bridge (explicit) | `scripts/quest_claude_bridge.py` | `claude --print`, API-metered after June 15, 2026 | explicit user/config opt-in, CI/daemonless/`ANTHROPIC_API_KEY` contexts |
 
 - `.ai/allowlist.json` `claude_role_transport` (`auto` | `background-agent` | `bridge`) is copied into `.quest/<id>/orchestration.json`; preflight records `claude_transport_resolved` there and keeps `claude_transport_downgraded: false` as a compatibility field. New `auto` runs stop for a user decision when bg fails; bridge is used only when explicitly selected/configured.
-- Native Claude-led Quest behavior is unchanged: Claude-designated roles still use native `Task(...)` execution when the orchestrator supports it.
+- Claude-designated roles use the Quest Claude runner under either orchestrator, with saved model/effort and the existing role persona. Background setup must pass preflight; native tasks are not a fallback.
 - The preferred helpers for Codex-led Claude slots are `scripts/quest_claude_probe.py --transport <t>` for transport preflight and `scripts/quest_claude_runner.py --transport <t>` for real role execution; the runner uses `bypassPermissions`, adds explicit repo/quest access via `--add-dir`, polls `handoff.json`, and updates `context_health.log` with the `transport=` field (the quest end summary and celebration report it).
 - The workflow probes transport availability once per session, sweeps orphaned `quest-<id>-*` background sessions at start/resume (`scripts/quest_claude_bg_run.py --sweep`), routes Claude-designated slots by selected model/runtime, and logs runner-invoked Claude roles as `runtime=claude`.
 - Transport failures are explicit: timeout retries once, CLI/auth/daemon failures block immediately with remediation (see `docs/guides/quest_setup.md`), and malformed output/missing handoff retries once before text fallback or blocking.

.ai/schemas/allowlist.schema.json

@@ -3,7 +3,7 @@
   "title": "Quest Allowlist Configuration",
   "description": "Creator-managed permissions for quest orchestration.",
   "type": "object",
-  "required": ["version", "auto_approve_phases", "review_mode", "role_permissions", "gates"],
+  "required": ["version", "auto_approve_phases", "review_mode", "role_permissions", "gates", "effort"],
   "properties": {
     "version": { "type": "integer", "minimum": 1 },
     "description": { "type": "string" },
@@ -45,7 +45,15 @@
     "codex_reasoning_effort": {
       "type": "string",
       "enum": ["low", "medium", "high", "xhigh", "max", "ultra"],
-      "description": "Reasoning effort pinned for Codex roles in new quests. Must be supported by the selected model and dispatch surface."
+      "description": "Default effort for standalone /gpt only. Quest roles use the per-role effort map."
+    },
+    "effort": {
+      "type": "object",
+      "description": "Saved per Quest role. Required for active Claude/Codex roles; omit for Gemini roles.",
+      "propertyNames": {
+        "enum": ["planner", "plan-reviewer-a", "plan-reviewer-b", "arbiter", "builder", "code-reviewer-a", "code-reviewer-b", "review-arbiter", "fixer"]
+      },
+      "additionalProperties": {"type": "string", "enum": ["low", "medium", "high", "xhigh", "max", "ultra"]}
     },
     "models": {
       "type": "object",

.skills/quest/agents/arbiter.md

@@ -4,7 +4,7 @@
 Gatekeeper for plan quality and canonical review decisions. Receives both plan-review artifacts, synthesizes their feedback, filters out noise, decides whether the plan is ready for implementation, and emits canonical findings artifacts.
 
 ## Tool
-Runtime is derived from `models.arbiter` in `.quest/<id>/orchestration.json`; the entrypoint (local Codex subagent, Quest Codex CLI runner, native `Task(...)`, or the bridge runner) follows the canonical dispatch matrix in `.skills/quest/delegation/workflow.md` (Runtime And Entrypoint Selection). That matrix is the single source of truth — do not restate or override it here.
+Runtime is derived from `models.arbiter` in `.quest/<id>/orchestration.json`; the entrypoint follows the canonical dispatch matrix in `.skills/quest/delegation/workflow.md` (Runtime And Entrypoint Selection). That matrix is the single source of truth — do not restate or override it here.
 
 ## Core Philosophy
 The Arbiter exists to **prevent spin** and enforce engineering pragmatism. It filters feedback through:

.skills/quest/agents/builder.md

@@ -4,7 +4,7 @@
 Implements the approved plan. Writes code, runs tests, produces a PR description.
 
 ## Tool
-Runtime is derived from `models.builder` in `.quest/<id>/orchestration.json`; the entrypoint (local Codex subagent, Quest Codex CLI runner, native `Task(...)`, or the bridge runner) follows the canonical dispatch matrix in `.skills/quest/delegation/workflow.md` (Runtime And Entrypoint Selection). That matrix is the single source of truth — do not restate or override it here.
+Runtime is derived from `models.builder` in `.quest/<id>/orchestration.json`; the entrypoint follows the canonical dispatch matrix in `.skills/quest/delegation/workflow.md` (Runtime And Entrypoint Selection). That matrix is the single source of truth — do not restate or override it here.
 
 When running on Codex, this role is non-interactive:
 - Do not ask questions.

.skills/quest/agents/fixer.md

@@ -6,7 +6,7 @@ At activation, announce the role and scope in one line. Example: `[fixer] fixing
 Fixes issues identified by the Code Review Agent. Applies targeted fixes and re-runs tests.
 
 ## Tool
-Runtime is derived from `models.fixer` in `.quest/<id>/orchestration.json`; the entrypoint (local Codex subagent, Quest Codex CLI runner, native `Task(...)`, or the bridge runner) follows the canonical dispatch matrix in `.skills/quest/delegation/workflow.md` (Runtime And Entrypoint Selection). That matrix is the single source of truth — do not restate or override it here.
+Runtime is derived from `models.fixer` in `.quest/<id>/orchestration.json`; the entrypoint follows the canonical dispatch matrix in `.skills/quest/delegation/workflow.md` (Runtime And Entrypoint Selection). That matrix is the single source of truth — do not restate or override it here.
 
 When running on Codex, this role is non-interactive:
 - Do not ask questions.

.skills/quest/agents/planner.md

@@ -4,7 +4,7 @@
 Creates and refines implementation plans from quest briefs. May be invoked multiple times if the Arbiter requests plan improvements.
 
 ## Tool
-Runtime is derived from `models.planner` in `.quest/<id>/orchestration.json`; the entrypoint (local Codex subagent, Quest Codex CLI runner, native `Task(...)`, or the bridge runner) follows the canonical dispatch matrix in `.skills/quest/delegation/workflow.md` (Runtime And Entrypoint Selection). That matrix is the single source of truth — do not restate or override it here.
+Runtime is derived from `models.planner` in `.quest/<id>/orchestration.json`; the entrypoint follows the canonical dispatch matrix in `.skills/quest/delegation/workflow.md` (Runtime And Entrypoint Selection). That matrix is the single source of truth — do not restate or override it here.
 
 When running on Codex, this role is non-interactive:
 - Do not ask questions.