feat(role): add versioned role configs + escalation ladder + scheduler (Phase 5)

Two parts:

Part A (fixes a gap from Phase 1): Groq/OpenRouter/OpenAI were documented in
docs/api-keys-setup.md as usable once configured, but nothing actually
constructed runners for them. internal/cli/cloudrunners.go consolidates
anthropic/google/groq/openrouter/openai NativeRunner construction into one
table-driven registerCloudRunners() helper, replacing the two hand-written
per-provider blocks in serve.go/run.go. Groq/OpenRouter/OpenAI reuse
openaicompat (no new adapter code) at SandboxKind: "docker".

Part B: the token-husbanding harness's core routing mechanism.

- internal/role: RoleConfig/Tier/Rung -- a role's system prompt and a
  multi-tier (provider, model) escalation ladder, versioned via config_json.
- storage: new role_configs table (draft/active/retired, UNIQUE(role,
  version)) with transactional activate-retires-prior-active semantics;
  new executions.escalation_rung column.
- task.AgentConfig.Role string -- purely additive; every existing task shape
  (Agent.Role == "") is unaffected, proven by TestPool_Execute_NonRoleTask_Unaffected
  plus the full pre-existing suite passing unchanged.
- executor.Pool.execute(): role-typed tasks with no Agent.Type yet resolve
  tier 0 of their active ladder (round-robin across multi-candidate tiers,
  skipping rate-limited providers, falling back to soonest-clearing) before
  the existing pickAgent/Classifier path runs; SystemPrompt applies to
  Agent.SystemPromptAppend. Already-resolved role tasks (scheduler resubmits)
  get their escalation_rung re-derived read-only via findTierIndex.
- internal/scheduler: polls role-typed FAILED tasks, retries at the same
  rung under MaxRetries or escalates to the next tier's first candidate when
  budget.Accountant.Allow() permits (emitting event.KindEscalated), else
  leaves the task FAILED with a final:true KindEscalated event. An
  in-memory per-execution-ID "handled" set keeps the poll loop convergent.
  Started by `serve` only, config knob [scheduler].poll_interval_seconds.
- internal/api: POST/GET /api/roles/{role}/versions, POST
  /api/roles/{role}/activate -- unauthenticated, matching the existing
  projects/tasks REST endpoints' auth posture (only chatbot MCP, agent MCP,
  and WebSocket are api_token-gated in this codebase today).

Documented as stored-but-not-yet-enforced (CLAUDE.md Design Debt, matching
how task.Priority/RetryConfig are already documented): RoleConfig.Tools/
SandboxKind don't affect dispatch yet; DefaultBudgetUSD is read narrowly as
the scheduler's escalation cost estimate, not enforced at initial dispatch;
scheduler escalation always targets Candidates[0] (no round-robin, unlike
initial-dispatch tier-0 resolution); the scheduler's dedupe is per-process
and resets on restart (idempotent, harmless).

go build/vet/test -race -count=1 all pass, 21 packages.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01V1moSNCJRcP6kykA4tyUSs

1 files changed, 147 insertions, 0 deletions

diff --git a/internal/storage/roleconfig.go b/internal/storage/roleconfig.go
new file mode 100644
index 0000000..6ae043b
--- /dev/null
+++ b/internal/storage/roleconfig.go
@@ -0,0 +1,147 @@
+package storage
+
+import (
+	"database/sql"
+	"fmt"
+	"time"
+
+	"github.com/google/uuid"
+)
+
+// RoleConfigRow is one versioned row in the role_configs table. ConfigJSON
+// holds the raw JSON-encoded role.RoleConfig; storage intentionally does not
+// import internal/role (no need to — it never inspects the payload, only
+// stores/retrieves it), so callers (internal/executor, internal/scheduler,
+// internal/api) decode it themselves via encoding/json.
+type RoleConfigRow struct {
+	ID          string
+	Role        string
+	Version     int
+	Status      string // "draft" | "active" | "retired"
+	ConfigJSON  string
+	CreatedAt   time.Time
+	ActivatedAt *time.Time
+	RetiredAt   *time.Time
+	ProposedBy  string
+}
+
+// CreateRoleConfig inserts a new draft version for role, auto-assigning the
+// next version number for that role (1 if none exist yet).
+func (s *DB) CreateRoleConfig(roleName, configJSON, proposedBy string) (*RoleConfigRow, error) {
+	tx, err := s.db.Begin()
+	if err != nil {
+		return nil, err
+	}
+	defer tx.Rollback() //nolint:errcheck
+
+	var maxVersion sql.NullInt64
+	if err := tx.QueryRow(`SELECT MAX(version) FROM role_configs WHERE role = ?`, roleName).Scan(&maxVersion); err != nil {
+		return nil, err
+	}
+	version := int(maxVersion.Int64) + 1
+
+	id := uuid.NewString()
+	now := time.Now().UTC()
+	if _, err := tx.Exec(`
+		INSERT INTO role_configs (id, role, version, status, config_json, created_at, proposed_by)
+		VALUES (?, ?, ?, 'draft', ?, ?, ?)`,
+		id, roleName, version, configJSON, now, proposedBy,
+	); err != nil {
+		return nil, err
+	}
+	if err := tx.Commit(); err != nil {
+		return nil, err
+	}
+
+	return &RoleConfigRow{
+		ID:         id,
+		Role:       roleName,
+		Version:    version,
+		Status:     "draft",
+		ConfigJSON: configJSON,
+		CreatedAt:  now,
+		ProposedBy: proposedBy,
+	}, nil
+}
+
+// GetActiveRoleConfig returns the currently active role_configs row for
+// role. Returns sql.ErrNoRows (unwrapped, matching GetTask/GetProject
+// convention in this package) if no version is active.
+func (s *DB) GetActiveRoleConfig(roleName string) (*RoleConfigRow, error) {
+	row := s.db.QueryRow(`
+		SELECT id, role, version, status, config_json, created_at, activated_at, retired_at, proposed_by
+		FROM role_configs WHERE role = ? AND status = 'active' LIMIT 1`, roleName)
+	return scanRoleConfigRow(row)
+}
+
+// ListRoleConfigVersions returns all versions for role, oldest first.
+func (s *DB) ListRoleConfigVersions(roleName string) ([]*RoleConfigRow, error) {
+	rows, err := s.db.Query(`
+		SELECT id, role, version, status, config_json, created_at, activated_at, retired_at, proposed_by
+		FROM role_configs WHERE role = ? ORDER BY version ASC`, roleName)
+	if err != nil {
+		return nil, err
+	}
+	defer rows.Close()
+
+	var out []*RoleConfigRow
+	for rows.Next() {
+		r, err := scanRoleConfigRow(rows)
+		if err != nil {
+			return nil, err
+		}
+		out = append(out, r)
+	}
+	return out, rows.Err()
+}
+
+// ActivateRoleConfigVersion promotes the given version of role to active,
+// atomically retiring whatever version is currently active for that role (if
+// any) in the same transaction. This enforces "at most one active row per
+// role" without a DB-level constraint, the same way UpdateTaskState enforces
+// the task state machine in a transaction rather than in schema.
+func (s *DB) ActivateRoleConfigVersion(roleName string, version int) error {
+	tx, err := s.db.Begin()
+	if err != nil {
+		return err
+	}
+	defer tx.Rollback() //nolint:errcheck
+
+	var exists int
+	if err := tx.QueryRow(`SELECT COUNT(*) FROM role_configs WHERE role = ? AND version = ?`, roleName, version).Scan(&exists); err != nil {
+		return err
+	}
+	if exists == 0 {
+		return fmt.Errorf("role %q version %d not found", roleName, version)
+	}
+
+	now := time.Now().UTC()
+	if _, err := tx.Exec(`UPDATE role_configs SET status = 'retired', retired_at = ? WHERE role = ? AND status = 'active'`, now, roleName); err != nil {
+		return err
+	}
+	if _, err := tx.Exec(`UPDATE role_configs SET status = 'active', activated_at = ? WHERE role = ? AND version = ?`, now, roleName, version); err != nil {
+		return err
+	}
+	return tx.Commit()
+}
+
+func scanRoleConfigRow(row scanner) (*RoleConfigRow, error) {
+	var r RoleConfigRow
+	var createdAt time.Time
+	var activatedAt, retiredAt sql.NullTime
+	var proposedBy sql.NullString
+	if err := row.Scan(&r.ID, &r.Role, &r.Version, &r.Status, &r.ConfigJSON, &createdAt, &activatedAt, &retiredAt, &proposedBy); err != nil {
+		return nil, err
+	}
+	r.CreatedAt = createdAt
+	if activatedAt.Valid {
+		t := activatedAt.Time
+		r.ActivatedAt = &t
+	}
+	if retiredAt.Valid {
+		t := retiredAt.Time
+		r.RetiredAt = &t
+	}
+	r.ProposedBy = proposedBy.String
+	return &r, nil
+}