summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorsoryu <soryu@soryu.co>2026-02-07 18:27:54 +0000
committersoryu <soryu@soryu.co>2026-02-07 18:27:54 +0000
commit97e21c8296ec5f91912d56980ebf3b18a1ca3507 (patch)
tree3650e2eb62ab5b387006563ce64139aa7688da5f
parent8f757f561eeb397aaea70d7c10d41445cc5e50b5 (diff)
downloadsoryu-97e21c8296ec5f91912d56980ebf3b18a1ca3507.tar.gz
soryu-97e21c8296ec5f91912d56980ebf3b18a1ca3507.zip
Add directive monitor contracts
Diffstat
-rw-r--r--makima/frontend/src/components/directives/DirectiveContractsTab.tsx131
-rw-r--r--makima/frontend/src/components/directives/DirectiveDetail.tsx393
-rw-r--r--makima/frontend/src/components/directives/StepDiagram.tsx152
-rw-r--r--makima/frontend/src/lib/api.ts14
-rw-r--r--makima/migrations/20260208000000_add_monitoring_contract_id.sql2
-rw-r--r--makima/src/bin/makima.rs14
-rw-r--r--makima/src/daemon/api/directive.rs26
-rw-r--r--makima/src/daemon/cli/directive.rs20
-rw-r--r--makima/src/daemon/cli/mod.rs6
-rw-r--r--makima/src/daemon/skills/directive.md20
-rw-r--r--makima/src/db/models.rs86
-rw-r--r--makima/src/db/repository.rs214
-rw-r--r--makima/src/orchestration/directive.rs562
-rw-r--r--makima/src/server/handlers/directives.rs223
-rw-r--r--makima/src/server/mod.rs2
-rw-r--r--makima/src/server/openapi.rs18
16 files changed, 1739 insertions, 144 deletions
diff --git a/makima/frontend/src/components/directives/DirectiveContractsTab.tsx b/makima/frontend/src/components/directives/DirectiveContractsTab.tsx
new file mode 100644
index 0000000..59ebfc8
--- /dev/null
+++ b/makima/frontend/src/components/directives/DirectiveContractsTab.tsx
@@ -0,0 +1,131 @@
+import { useNavigate } from "react-router";
+import type {
+ DirectiveWithChains,
+ StepContractSummary,
+ ContractPhase,
+} from "../../lib/api";
+import { PhaseProgressBarCompact } from "../contracts/PhaseProgressBar";
+
+interface DirectiveContractsTabProps {
+ directive: DirectiveWithChains;
+}
+
+const statusColors: Record<string, string> = {
+ active: "text-green-400",
+ completed: "text-blue-400",
+ archived: "text-[#555]",
+};
+
+function ContractCard({
+ summary,
+ label,
+}: {
+ summary: StepContractSummary;
+ label: string;
+}) {
+ const navigate = useNavigate();
+
+ const progressPct =
+ summary.taskCount > 0
+ ? Math.round((summary.tasksDone / summary.taskCount) * 100)
+ : 0;
+
+ return (
+ <div
+ className="border border-dashed border-[rgba(117,170,252,0.25)] bg-[rgba(117,170,252,0.03)] p-3 cursor-pointer hover:bg-[rgba(117,170,252,0.06)] transition-colors"
+ onClick={() => navigate(`/contracts/${summary.id}`)}
+ >
+ <div className="flex items-center justify-between mb-1.5">
+ <div className="flex items-center gap-2 min-w-0">
+ <span className="font-mono text-[11px] text-[#dbe7ff] truncate">
+ {summary.name}
+ </span>
+ <span className="font-mono text-[9px] text-[#7788aa] uppercase shrink-0">
+ {summary.contractType}
+ </span>
+ </div>
+ <div className="flex items-center gap-2 shrink-0">
+ <span
+ className={`font-mono text-[9px] uppercase ${statusColors[summary.status] || "text-[#888]"}`}
+ >
+ {summary.status}
+ </span>
+ <span className="font-mono text-[9px] text-[#75aafc]">&rarr;</span>
+ </div>
+ </div>
+
+ <div className="flex items-center gap-2 mb-1.5">
+ <span className="font-mono text-[9px] text-[#7788aa] uppercase shrink-0">
+ {label}
+ </span>
+ <PhaseProgressBarCompact
+ currentPhase={summary.phase as ContractPhase}
+ />
+ </div>
+
+ {/* Task progress bar */}
+ <div className="flex items-center gap-2">
+ <div className="flex-1 h-1 bg-[rgba(117,170,252,0.1)] rounded-full overflow-hidden">
+ <div
+ className="h-full bg-[#3f6fb3] rounded-full transition-all"
+ style={{ width: `${progressPct}%` }}
+ />
+ </div>
+ <span className="font-mono text-[9px] text-[#7788aa] shrink-0">
+ {summary.tasksDone}/{summary.taskCount} tasks
+ </span>
+ </div>
+ </div>
+ );
+}
+
+export function DirectiveContractsTab({
+ directive,
+}: DirectiveContractsTabProps) {
+ // Collect all contract summaries
+ const contracts: { summary: StepContractSummary; label: string }[] = [];
+
+ if (directive.orchestratorContractSummary) {
+ contracts.push({
+ summary: directive.orchestratorContractSummary,
+ label: "Planning",
+ });
+ }
+
+ for (const chain of directive.chains) {
+ for (const step of chain.steps) {
+ if (step.contractSummary) {
+ contracts.push({
+ summary: step.contractSummary,
+ label: step.name,
+ });
+ }
+ }
+ }
+
+ if (contracts.length === 0) {
+ return (
+ <div className="text-center py-8">
+ <p className="font-mono text-xs text-[#7788aa]">
+ {directive.status === "draft"
+ ? "No contracts yet. Start the directive to begin planning."
+ : directive.status === "planning"
+ ? "Planning in progress... contracts will appear when steps are created."
+ : "No contracts associated with this directive."}
+ </p>
+ </div>
+ );
+ }
+
+ return (
+ <div className="space-y-2">
+ {contracts.map((c) => (
+ <ContractCard
+ key={c.summary.id}
+ summary={c.summary}
+ label={c.label}
+ />
+ ))}
+ </div>
+ );
+}
diff --git a/makima/frontend/src/components/directives/DirectiveDetail.tsx b/makima/frontend/src/components/directives/DirectiveDetail.tsx
index 094cdf2..95dc7cc 100644
--- a/makima/frontend/src/components/directives/DirectiveDetail.tsx
+++ b/makima/frontend/src/components/directives/DirectiveDetail.tsx
@@ -1,12 +1,16 @@
-import { useEffect, useRef } from "react";
+import { useState, useEffect, useRef } from "react";
import { useNavigate } from "react-router";
import type {
DirectiveWithChains,
DirectiveStatus,
ChainWithSteps,
ChainStep,
+ ContractPhase,
} from "../../lib/api";
import { getDirective } from "../../lib/api";
+import { PhaseProgressBarCompact } from "../contracts/PhaseProgressBar";
+import { StepDiagram } from "./StepDiagram";
+import { DirectiveContractsTab } from "./DirectiveContractsTab";
interface DirectiveDetailProps {
directive: DirectiveWithChains;
@@ -16,6 +20,8 @@ interface DirectiveDetailProps {
onRefresh?: (updated: DirectiveWithChains) => void;
}
+type Tab = "overview" | "chain" | "contracts";
+
const statusColors: Record<DirectiveStatus, string> = {
draft: "text-[#888]",
planning: "text-yellow-400",
@@ -36,17 +42,18 @@ const stepStatusColors: Record<string, string> = {
const stepStatusIcons: Record<string, string> = {
pending: "\u25CB", // β—‹
running: "\u25D4", // β—”
- passed: "\u25CF", // ●
- failed: "\u2715", // βœ•
+ passed: "\u25CF", // ●
+ failed: "\u2715", // βœ•
};
function StepRow({ step }: { step: ChainStep }) {
const navigate = useNavigate();
const color = stepStatusColors[step.status] || "text-[#888]";
const icon = stepStatusIcons[step.status] || "\u25CB";
+ const summary = step.contractSummary;
return (
- <div className="flex items-start gap-2 py-1 px-2 hover:bg-[rgba(117,170,252,0.05)]">
+ <div className="flex items-start gap-2 py-1.5 px-2 hover:bg-[rgba(117,170,252,0.05)]">
<span className={`font-mono text-[11px] ${color} mt-px`}>{icon}</span>
<div className="flex-1 min-w-0">
<div className="flex items-center gap-2">
@@ -57,21 +64,41 @@ function StepRow({ step }: { step: ChainStep }) {
{step.status}
</span>
</div>
+ {summary && (
+ <div className="flex items-center gap-2 mt-0.5">
+ <PhaseProgressBarCompact
+ currentPhase={summary.phase as ContractPhase}
+ />
+ <span className="font-mono text-[9px] text-[#7788aa]">
+ {summary.tasksDone}/{summary.taskCount} tasks
+ </span>
+ {step.contractId && (
+ <button
+ onClick={(e) => {
+ e.stopPropagation();
+ navigate(`/contracts/${step.contractId}`);
+ }}
+ className="font-mono text-[9px] text-[#75aafc] hover:text-white transition-colors"
+ >
+ contract &rarr;
+ </button>
+ )}
+ </div>
+ )}
+ {!summary && step.contractId && (
+ <button
+ onClick={() => navigate(`/contracts/${step.contractId}`)}
+ className="font-mono text-[9px] text-[#75aafc] hover:text-white transition-colors mt-0.5"
+ >
+ contract &rarr;
+ </button>
+ )}
{step.description && (
- <p className="font-mono text-[10px] text-[#7788aa] truncate">
+ <p className="font-mono text-[10px] text-[#7788aa] truncate mt-0.5">
{step.description}
</p>
)}
</div>
- {step.contractId && (
- <button
- onClick={() => navigate(`/contracts/${step.contractId}`)}
- className="font-mono text-[9px] text-[#75aafc] hover:text-white transition-colors shrink-0"
- title="View contract"
- >
- contract &rarr;
- </button>
- )}
</div>
);
}
@@ -84,7 +111,9 @@ function ChainCard({ chainWithSteps }: { chainWithSteps: ChainWithSteps }) {
<div className="border border-dashed border-[rgba(117,170,252,0.25)] bg-[rgba(117,170,252,0.03)]">
<div className="p-3">
<div className="flex items-center justify-between mb-1">
- <span className="font-mono text-xs text-[#dbe7ff]">{chain.name}</span>
+ <span className="font-mono text-xs text-[#dbe7ff]">
+ {chain.name}
+ </span>
<span className="font-mono text-[10px] text-[#7788aa] uppercase">
gen {chain.generation} &middot; {chain.status}
</span>
@@ -102,7 +131,9 @@ function ChainCard({ chainWithSteps }: { chainWithSteps: ChainWithSteps }) {
<span className="text-red-400">{chain.failedSteps} failed</span>
)}
{chain.currentConfidence != null && (
- <span>confidence: {(chain.currentConfidence * 100).toFixed(0)}%</span>
+ <span>
+ confidence: {(chain.currentConfidence * 100).toFixed(0)}%
+ </span>
)}
</div>
</div>
@@ -151,6 +182,7 @@ export function DirectiveDetail({
onRefresh,
}: DirectiveDetailProps) {
const navigate = useNavigate();
+ const [activeTab, setActiveTab] = useState<Tab>("overview");
// Auto-poll when directive is in an active state
const isLive =
@@ -185,6 +217,31 @@ export function DirectiveDetail({
};
}, [isLive, directive.id, onRefresh]);
+ // Count total steps and completed steps across all chains
+ const totalSteps = directive.chains.reduce(
+ (sum, c) => sum + c.totalSteps,
+ 0
+ );
+ const completedSteps = directive.chains.reduce(
+ (sum, c) => sum + c.completedSteps,
+ 0
+ );
+
+ // Count contracts
+ const contractCount =
+ (directive.orchestratorContractSummary ? 1 : 0) +
+ directive.chains.reduce(
+ (sum, c) =>
+ sum + c.steps.filter((s) => s.contractSummary != null).length,
+ 0
+ );
+
+ const tabs: { key: Tab; label: string; count?: number }[] = [
+ { key: "overview", label: "Overview" },
+ { key: "chain", label: "Chain", count: totalSteps },
+ { key: "contracts", label: "Contracts", count: contractCount },
+ ];
+
return (
<div className="panel h-full flex flex-col">
{/* Header */}
@@ -235,109 +292,235 @@ export function DirectiveDetail({
</h2>
</div>
- {/* Content */}
- <div className="flex-1 overflow-y-auto p-4 space-y-4">
- {/* Orchestrator contract link */}
- {directive.orchestratorContractId && (
- <div className="flex items-center gap-2 p-2 border border-dashed border-[rgba(117,170,252,0.2)] bg-[rgba(117,170,252,0.03)]">
- <span className="font-mono text-[10px] text-[#7788aa] uppercase">
- Planning Contract
- </span>
- <button
- onClick={() =>
- navigate(`/contracts/${directive.orchestratorContractId}`)
+ {/* Tabs */}
+ <div className="flex border-b border-[rgba(117,170,252,0.2)]">
+ {tabs.map((tab) => (
+ <button
+ key={tab.key}
+ onClick={() => setActiveTab(tab.key)}
+ className={`
+ px-4 py-2 font-mono text-xs uppercase tracking-wider transition-colors
+ ${
+ activeTab === tab.key
+ ? "text-[#dbe7ff] border-b-2 border-[#75aafc]"
+ : "text-[#555] hover:text-[#9bc3ff]"
}
- className="font-mono text-[11px] text-[#75aafc] hover:text-white transition-colors"
- >
- {directive.orchestratorContractId.slice(0, 8)}... &rarr;
- </button>
- {directive.status === "planning" && (
- <span className="font-mono text-[9px] text-yellow-400 animate-pulse">
- planning in progress
+ `}
+ >
+ {tab.label}
+ {tab.count != null && tab.count > 0 && (
+ <span className="ml-1 text-[10px] text-[#7788aa]">
+ ({tab.count})
</span>
)}
- </div>
- )}
+ </button>
+ ))}
+ </div>
- {/* Goal */}
- <div>
- <h4 className="font-mono text-[10px] text-[#75aafc] uppercase tracking-wider mb-1">
- Goal
- </h4>
- <p className="font-mono text-xs text-[#9bb8d8] whitespace-pre-wrap">
- {directive.goal}
- </p>
- </div>
+ {/* Tab content */}
+ <div className="flex-1 overflow-y-auto p-4">
+ {activeTab === "overview" && (
+ <div className="space-y-4">
+ {/* Orchestrator contract link */}
+ {directive.orchestratorContractId && (
+ <div className="flex items-center gap-2 p-2 border border-dashed border-[rgba(117,170,252,0.2)] bg-[rgba(117,170,252,0.03)]">
+ <span className="font-mono text-[10px] text-[#7788aa] uppercase">
+ Planning Contract
+ </span>
+ {directive.orchestratorContractSummary && (
+ <PhaseProgressBarCompact
+ currentPhase={
+ directive.orchestratorContractSummary
+ .phase as ContractPhase
+ }
+ />
+ )}
+ <button
+ onClick={() =>
+ navigate(
+ `/contracts/${directive.orchestratorContractId}`
+ )
+ }
+ className="font-mono text-[11px] text-[#75aafc] hover:text-white transition-colors"
+ >
+ {directive.orchestratorContractSummary?.name ||
+ directive.orchestratorContractId.slice(0, 8) + "..."}{" "}
+ &rarr;
+ </button>
+ {directive.status === "planning" && (
+ <span className="font-mono text-[9px] text-yellow-400 animate-pulse">
+ planning in progress
+ </span>
+ )}
+ </div>
+ )}
- {/* Config */}
- <div className="grid grid-cols-2 gap-2">
- <div>
- <span className="font-mono text-[10px] text-[#7788aa] uppercase">
- Autonomy
- </span>
- <div className="font-mono text-xs text-[#dbe7ff]">
- {directive.autonomyLevel}
+ {/* Goal */}
+ <div>
+ <h4 className="font-mono text-[10px] text-[#75aafc] uppercase tracking-wider mb-1">
+ Goal
+ </h4>
+ <p className="font-mono text-xs text-[#9bb8d8] whitespace-pre-wrap">
+ {directive.goal}
+ </p>
</div>
- </div>
- <div>
- <span className="font-mono text-[10px] text-[#7788aa] uppercase">
- Chains
- </span>
- <div className="font-mono text-xs text-[#dbe7ff]">
- {directive.chainGenerationCount} generated
+
+ {/* Config grid */}
+ <div className="grid grid-cols-3 gap-2">
+ <div>
+ <span className="font-mono text-[10px] text-[#7788aa] uppercase">
+ Autonomy
+ </span>
+ <div className="font-mono text-xs text-[#dbe7ff]">
+ {directive.autonomyLevel}
+ </div>
+ </div>
+ <div>
+ <span className="font-mono text-[10px] text-[#7788aa] uppercase">
+ Chains
+ </span>
+ <div className="font-mono text-xs text-[#dbe7ff]">
+ {directive.chainGenerationCount} generated
+ </div>
+ </div>
+ <div>
+ <span className="font-mono text-[10px] text-[#7788aa] uppercase">
+ Cost
+ </span>
+ <div className="font-mono text-xs text-[#dbe7ff]">
+ ${directive.totalCostUsd.toFixed(2)}
+ </div>
+ </div>
+ {directive.repositoryUrl && (
+ <div className="col-span-3">
+ <span className="font-mono text-[10px] text-[#7788aa] uppercase">
+ Repository
+ </span>
+ <div className="font-mono text-xs text-[#dbe7ff] truncate">
+ {directive.repositoryUrl}
+ </div>
+ </div>
+ )}
</div>
- </div>
- <div>
- <span className="font-mono text-[10px] text-[#7788aa] uppercase">
- Cost
- </span>
- <div className="font-mono text-xs text-[#dbe7ff]">
- ${directive.totalCostUsd.toFixed(2)}
+
+ {/* Stat cards */}
+ <div className="grid grid-cols-3 gap-2">
+ <div className="border border-dashed border-[rgba(117,170,252,0.2)] p-2 text-center">
+ <div className="font-mono text-lg text-[#dbe7ff]">
+ {totalSteps}
+ </div>
+ <div className="font-mono text-[9px] text-[#7788aa] uppercase">
+ Total Steps
+ </div>
+ </div>
+ <div className="border border-dashed border-[rgba(117,170,252,0.2)] p-2 text-center">
+ <div className="font-mono text-lg text-green-400">
+ {completedSteps}
+ </div>
+ <div className="font-mono text-[9px] text-[#7788aa] uppercase">
+ Completed
+ </div>
+ </div>
+ <div className="border border-dashed border-[rgba(117,170,252,0.2)] p-2 text-center">
+ <div className="font-mono text-lg text-[#dbe7ff]">
+ ${directive.totalCostUsd.toFixed(2)}
+ </div>
+ <div className="font-mono text-[9px] text-[#7788aa] uppercase">
+ Cost
+ </div>
+ </div>
</div>
- </div>
- {directive.repositoryUrl && (
+
+ {/* Structured sections */}
+ <JsonSection label="Requirements" data={directive.requirements} />
+ <JsonSection
+ label="Acceptance Criteria"
+ data={directive.acceptanceCriteria}
+ />
+ <JsonSection label="Constraints" data={directive.constraints} />
+ <JsonSection
+ label="External Dependencies"
+ data={directive.externalDependencies}
+ />
+
+ {/* Metadata */}
<div>
- <span className="font-mono text-[10px] text-[#7788aa] uppercase">
- Repository
- </span>
- <div className="font-mono text-xs text-[#dbe7ff] truncate">
- {directive.repositoryUrl}
+ <h4 className="font-mono text-[10px] text-[#75aafc] uppercase tracking-wider mb-1">
+ Metadata
+ </h4>
+ <div className="grid grid-cols-2 gap-1 font-mono text-[10px]">
+ <span className="text-[#7788aa]">Created</span>
+ <span className="text-[#9bb8d8]">
+ {new Date(directive.createdAt).toLocaleString()}
+ </span>
+ <span className="text-[#7788aa]">Updated</span>
+ <span className="text-[#9bb8d8]">
+ {new Date(directive.updatedAt).toLocaleString()}
+ </span>
+ {directive.startedAt && (
+ <>
+ <span className="text-[#7788aa]">Started</span>
+ <span className="text-[#9bb8d8]">
+ {new Date(directive.startedAt).toLocaleString()}
+ </span>
+ </>
+ )}
+ {directive.completedAt && (
+ <>
+ <span className="text-[#7788aa]">Completed</span>
+ <span className="text-[#9bb8d8]">
+ {new Date(directive.completedAt).toLocaleString()}
+ </span>
+ </>
+ )}
+ <span className="text-[#7788aa]">Version</span>
+ <span className="text-[#9bb8d8]">{directive.version}</span>
</div>
</div>
- )}
- </div>
+ </div>
+ )}
- {/* Structured sections */}
- <JsonSection label="Requirements" data={directive.requirements} />
- <JsonSection
- label="Acceptance Criteria"
- data={directive.acceptanceCriteria}
- />
- <JsonSection label="Constraints" data={directive.constraints} />
- <JsonSection
- label="External Dependencies"
- data={directive.externalDependencies}
- />
+ {activeTab === "chain" && (
+ <div className="space-y-4">
+ {/* Step diagram */}
+ {directive.chains.length > 0 && (
+ <div>
+ <h4 className="font-mono text-[10px] text-[#75aafc] uppercase tracking-wider mb-2">
+ Step Dependencies
+ </h4>
+ <StepDiagram
+ steps={directive.chains.flatMap((c) => c.steps)}
+ />
+ </div>
+ )}
- {/* Chains */}
- <div>
- <h4 className="font-mono text-[10px] text-[#75aafc] uppercase tracking-wider mb-2">
- Chains ({directive.chains.length})
- </h4>
- {directive.chains.length === 0 ? (
- <p className="font-mono text-xs text-[#7788aa]">
- {directive.status === "planning"
- ? "Planning in progress... chains will appear when the planner completes."
- : "No chains yet. Chains are created during planning."}
- </p>
- ) : (
- <div className="space-y-2">
- {directive.chains.map((cws) => (
- <ChainCard key={cws.id} chainWithSteps={cws} />
- ))}
+ {/* Chain cards */}
+ <div>
+ <h4 className="font-mono text-[10px] text-[#75aafc] uppercase tracking-wider mb-2">
+ Chains ({directive.chains.length})
+ </h4>
+ {directive.chains.length === 0 ? (
+ <p className="font-mono text-xs text-[#7788aa]">
+ {directive.status === "planning"
+ ? "Planning in progress... chains will appear when the planner completes."
+ : directive.status === "draft"
+ ? "No chains yet. Start the directive to begin planning."
+ : "No chains created for this directive."}
+ </p>
+ ) : (
+ <div className="space-y-2">
+ {directive.chains.map((cws) => (
+ <ChainCard key={cws.id} chainWithSteps={cws} />
+ ))}
+ </div>
+ )}
</div>
- )}
- </div>
+ </div>
+ )}
+
+ {activeTab === "contracts" && (
+ <DirectiveContractsTab directive={directive} />
+ )}
</div>
</div>
);
diff --git a/makima/frontend/src/components/directives/StepDiagram.tsx b/makima/frontend/src/components/directives/StepDiagram.tsx
new file mode 100644
index 0000000..5c65ae1
--- /dev/null
+++ b/makima/frontend/src/components/directives/StepDiagram.tsx
@@ -0,0 +1,152 @@
+import { useNavigate } from "react-router";
+import type { ChainStep, ContractPhase } from "../../lib/api";
+import { PhaseProgressBarCompact } from "../contracts/PhaseProgressBar";
+
+interface StepDiagramProps {
+ steps: ChainStep[];
+}
+
+const statusBorderColors: Record<string, string> = {
+ pending: "border-[#555]",
+ running: "border-yellow-400",
+ passed: "border-green-400",
+ failed: "border-red-400",
+};
+
+const statusDotColors: Record<string, string> = {
+ pending: "bg-[#555]",
+ running: "bg-yellow-400",
+ passed: "bg-green-400",
+ failed: "bg-red-400",
+};
+
+/**
+ * Assign depth to each step via topological sort.
+ * Steps with no dependsOn = depth 0. Steps depending only on depth-0 = depth 1. Etc.
+ */
+function assignDepths(steps: ChainStep[]): Map<string, number> {
+ const depths = new Map<string, number>();
+ const stepMap = new Map(steps.map((s) => [s.id, s]));
+
+ function getDepth(id: string): number {
+ if (depths.has(id)) return depths.get(id)!;
+ const step = stepMap.get(id);
+ if (!step || !step.dependsOn || step.dependsOn.length === 0) {
+ depths.set(id, 0);
+ return 0;
+ }
+ const maxParent = Math.max(
+ ...step.dependsOn.map((depId) => getDepth(depId))
+ );
+ const d = maxParent + 1;
+ depths.set(id, d);
+ return d;
+ }
+
+ for (const step of steps) {
+ getDepth(step.id);
+ }
+
+ return depths;
+}
+
+export function StepDiagram({ steps }: StepDiagramProps) {
+ const navigate = useNavigate();
+
+ if (steps.length === 0) {
+ return (
+ <p className="font-mono text-xs text-[#7788aa]">No steps to display.</p>
+ );
+ }
+
+ const depths = assignDepths(steps);
+ const maxDepth = Math.max(...Array.from(depths.values()));
+
+ // Group steps by depth
+ const levels: ChainStep[][] = [];
+ for (let d = 0; d <= maxDepth; d++) {
+ levels.push(
+ steps
+ .filter((s) => depths.get(s.id) === d)
+ .sort((a, b) => a.orderIndex - b.orderIndex)
+ );
+ }
+
+ // Build position map for connectors
+ const stepPositions = new Map<string, { level: number; index: number }>();
+ levels.forEach((level, li) => {
+ level.forEach((step, si) => {
+ stepPositions.set(step.id, { level: li, index: si });
+ });
+ });
+
+ return (
+ <div className="space-y-3">
+ {levels.map((level, li) => (
+ <div key={li} className="flex items-start gap-2 flex-wrap">
+ {li > 0 && (
+ <div className="w-full flex justify-center mb-1">
+ <div className="w-px h-3 bg-[rgba(117,170,252,0.3)]" />
+ </div>
+ )}
+ {level.map((step) => {
+ const borderColor =
+ statusBorderColors[step.status] || "border-[#555]";
+ const dotColor = statusDotColors[step.status] || "bg-[#555]";
+ const summary = step.contractSummary;
+ const hasContract = !!step.contractId;
+
+ return (
+ <div
+ key={step.id}
+ className={`
+ border ${borderColor} bg-[rgba(0,0,0,0.2)] p-2 min-w-[180px] max-w-[220px]
+ ${hasContract ? "cursor-pointer hover:bg-[rgba(117,170,252,0.05)]" : ""}
+ transition-colors
+ `}
+ onClick={() => {
+ if (hasContract) navigate(`/contracts/${step.contractId}`);
+ }}
+ title={hasContract ? "View contract" : undefined}
+ >
+ <div className="flex items-center gap-1.5 mb-1">
+ <div className={`w-1.5 h-1.5 rounded-full ${dotColor}`} />
+ <span className="font-mono text-[11px] text-[#dbe7ff] truncate flex-1">
+ {step.name}
+ </span>
+ {hasContract && (
+ <span className="font-mono text-[9px] text-[#75aafc] shrink-0">
+ &rarr;
+ </span>
+ )}
+ </div>
+ {summary && (
+ <>
+ <div className="mb-1">
+ <PhaseProgressBarCompact
+ currentPhase={summary.phase as ContractPhase}
+ />
+ </div>
+ <div className="font-mono text-[9px] text-[#7788aa]">
+ {summary.tasksDone}/{summary.taskCount} tasks
+ {summary.tasksRunning > 0 && (
+ <span className="text-yellow-400 ml-1">
+ {summary.tasksRunning} running
+ </span>
+ )}
+ {summary.tasksFailed > 0 && (
+ <span className="text-red-400 ml-1">
+ {summary.tasksFailed} failed
+ </span>
+ )}
+ </div>
+ </>
+ )}
+ </div>
+ );
+ })}
+ </div>
+ ))}
+ </div>
+ );
+}
diff --git a/makima/frontend/src/lib/api.ts b/makima/frontend/src/lib/api.ts
index ccc7156..9782a07 100644
--- a/makima/frontend/src/lib/api.ts
+++ b/makima/frontend/src/lib/api.ts
@@ -3075,6 +3075,18 @@ export interface DirectiveChain {
updatedAt: string;
}
+export interface StepContractSummary {
+ id: string;
+ name: string;
+ contractType: string;
+ phase: string;
+ status: string;
+ taskCount: number;
+ tasksDone: number;
+ tasksRunning: number;
+ tasksFailed: number;
+}
+
export interface ChainStep {
id: string;
chainId: string;
@@ -3092,6 +3104,7 @@ export interface ChainStep {
startedAt: string | null;
completedAt: string | null;
createdAt: string;
+ contractSummary: StepContractSummary | null;
}
export interface ChainWithSteps extends DirectiveChain {
@@ -3099,6 +3112,7 @@ export interface ChainWithSteps extends DirectiveChain {
}
export interface DirectiveWithChains extends Directive {
+ orchestratorContractSummary: StepContractSummary | null;
chains: ChainWithSteps[];
}
diff --git a/makima/migrations/20260208000000_add_monitoring_contract_id.sql b/makima/migrations/20260208000000_add_monitoring_contract_id.sql
new file mode 100644
index 0000000..65a4a0d
--- /dev/null
+++ b/makima/migrations/20260208000000_add_monitoring_contract_id.sql
@@ -0,0 +1,2 @@
+ALTER TABLE chain_steps ADD COLUMN monitoring_contract_id UUID REFERENCES contracts(id) ON DELETE SET NULL;
+ALTER TABLE chain_steps ADD COLUMN monitoring_task_id UUID;
diff --git a/makima/src/bin/makima.rs b/makima/src/bin/makima.rs
index 9d7f847..92fdae6 100644
--- a/makima/src/bin/makima.rs
+++ b/makima/src/bin/makima.rs
@@ -773,6 +773,20 @@ async fn run_directive(
let result = client.directive_start(args.directive_id).await?;
println!("{}", serde_json::to_string(&result.0)?);
}
+ DirectiveCommand::Evaluate(args) => {
+ let client = ApiClient::new(args.common.api_url, args.common.api_key)?;
+ let result = client
+ .directive_evaluate_step(args.common.directive_id, args.step_id)
+ .await?;
+ println!("{}", serde_json::to_string(&result.0)?);
+ }
+ DirectiveCommand::Evaluations(args) => {
+ let client = ApiClient::new(args.common.api_url, args.common.api_key)?;
+ let result = client
+ .directive_evaluations(args.common.directive_id, args.step_id)
+ .await?;
+ println!("{}", serde_json::to_string(&result.0)?);
+ }
}
Ok(())
diff --git a/makima/src/daemon/api/directive.rs b/makima/src/daemon/api/directive.rs
index 42f6f45..3589e78 100644
--- a/makima/src/daemon/api/directive.rs
+++ b/makima/src/daemon/api/directive.rs
@@ -57,4 +57,30 @@ impl ApiClient {
self.post_empty(&format!("/api/v1/directives/{}/start", directive_id))
.await
}
+
+ /// Trigger a manual evaluation for a step.
+ pub async fn directive_evaluate_step(
+ &self,
+ directive_id: Uuid,
+ step_id: Uuid,
+ ) -> Result<JsonValue, ApiError> {
+ self.post_empty(&format!(
+ "/api/v1/directives/{}/steps/{}/evaluate",
+ directive_id, step_id
+ ))
+ .await
+ }
+
+ /// List evaluations for a step.
+ pub async fn directive_evaluations(
+ &self,
+ directive_id: Uuid,
+ step_id: Uuid,
+ ) -> Result<JsonValue, ApiError> {
+ self.get(&format!(
+ "/api/v1/directives/{}/steps/{}/evaluations",
+ directive_id, step_id
+ ))
+ .await
+ }
}
diff --git a/makima/src/daemon/cli/directive.rs b/makima/src/daemon/cli/directive.rs
index 5ce88c5..4c29c14 100644
--- a/makima/src/daemon/cli/directive.rs
+++ b/makima/src/daemon/cli/directive.rs
@@ -38,3 +38,23 @@ pub struct UpdateStatusArgs {
/// New status (draft, planning, active, paused, completed, archived, failed)
pub status: String,
}
+
+/// Arguments for evaluate command (trigger manual evaluation).
+#[derive(Args, Debug)]
+pub struct EvaluateArgs {
+ #[command(flatten)]
+ pub common: DirectiveArgs,
+
+ /// Step ID to evaluate
+ pub step_id: Uuid,
+}
+
+/// Arguments for evaluations command (list evaluation history).
+#[derive(Args, Debug)]
+pub struct EvaluationsArgs {
+ #[command(flatten)]
+ pub common: DirectiveArgs,
+
+ /// Step ID to list evaluations for
+ pub step_id: Uuid,
+}
diff --git a/makima/src/daemon/cli/mod.rs b/makima/src/daemon/cli/mod.rs
index b07ab5a..c9a8c6f 100644
--- a/makima/src/daemon/cli/mod.rs
+++ b/makima/src/daemon/cli/mod.rs
@@ -225,6 +225,12 @@ pub enum DirectiveCommand {
/// Start a directive (create planning contract and begin orchestration)
Start(DirectiveArgs),
+
+ /// Trigger a manual evaluation for a step
+ Evaluate(directive::EvaluateArgs),
+
+ /// List evaluation history for a step
+ Evaluations(directive::EvaluationsArgs),
}
impl Cli {
diff --git a/makima/src/daemon/skills/directive.md b/makima/src/daemon/skills/directive.md
index cdfdaa2..0d1e9d6 100644
--- a/makima/src/daemon/skills/directive.md
+++ b/makima/src/daemon/skills/directive.md
@@ -47,6 +47,20 @@ makima directive update-status <status>
```
Updates the directive status. Valid statuses: `draft`, `planning`, `active`, `paused`, `completed`, `archived`, `failed`.
+## Evaluation
+
+### Trigger manual evaluation for a step
+```bash
+makima directive evaluate <step_id>
+```
+Triggers a monitoring evaluation for the specified step. The step must have been executed (have a contract). Sets the step to "evaluating" and dispatches a monitoring contract.
+
+### List evaluations for a step
+```bash
+makima directive evaluations <step_id>
+```
+Returns the evaluation history for a step, ordered by evaluation number.
+
## Output Format
All commands output JSON to stdout.
@@ -63,6 +77,12 @@ makima directive chains
# Get details of a specific chain
makima directive chain <chain_id>
+# Trigger manual evaluation of a step
+makima directive evaluate <step_id>
+
+# Check evaluation history
+makima directive evaluations <step_id>
+
# Update status to active
makima directive update-status active
```
diff --git a/makima/src/db/models.rs b/makima/src/db/models.rs
index bc90942..eff2df0 100644
--- a/makima/src/db/models.rs
+++ b/makima/src/db/models.rs
@@ -2849,12 +2849,37 @@ pub struct UpdateDirectiveRequest {
pub version: Option<i32>,
}
+/// Lightweight contract summary attached to a chain step.
+#[derive(Debug, FromRow, Serialize, ToSchema)]
+#[serde(rename_all = "camelCase")]
+pub struct StepContractSummary {
+ pub id: Uuid,
+ pub name: String,
+ pub contract_type: String,
+ pub phase: String,
+ pub status: String,
+ pub task_count: i64,
+ pub tasks_done: i64,
+ pub tasks_running: i64,
+ pub tasks_failed: i64,
+}
+
+/// Chain step enriched with optional contract summary.
+#[derive(Debug, Serialize, ToSchema)]
+#[serde(rename_all = "camelCase")]
+pub struct ChainStepWithContract {
+ #[serde(flatten)]
+ pub step: ChainStep,
+ pub contract_summary: Option<StepContractSummary>,
+}
+
/// Directive with its chains and steps for detail view.
#[derive(Debug, Serialize, ToSchema)]
#[serde(rename_all = "camelCase")]
pub struct DirectiveWithChains {
#[serde(flatten)]
pub directive: Directive,
+ pub orchestrator_contract_summary: Option<StepContractSummary>,
pub chains: Vec<ChainWithSteps>,
}
@@ -2903,6 +2928,8 @@ pub struct ChainStep {
pub status: String,
pub contract_id: Option<Uuid>,
pub supervisor_task_id: Option<Uuid>,
+ pub monitoring_contract_id: Option<Uuid>,
+ pub monitoring_task_id: Option<Uuid>,
pub confidence_score: Option<f64>,
pub confidence_level: Option<String>,
pub evaluation_count: i32,
@@ -2922,5 +2949,62 @@ pub struct ChainStep {
pub struct ChainWithSteps {
#[serde(flatten)]
pub chain: DirectiveChain,
- pub steps: Vec<ChainStep>,
+ pub steps: Vec<ChainStepWithContract>,
+}
+
+/// Full row from directive_evaluations table.
+#[derive(Debug, Clone, FromRow, Serialize, ToSchema)]
+#[serde(rename_all = "camelCase")]
+pub struct DirectiveEvaluation {
+ pub id: Uuid,
+ pub directive_id: Uuid,
+ pub chain_id: Option<Uuid>,
+ pub step_id: Option<Uuid>,
+ pub contract_id: Option<Uuid>,
+ pub evaluation_type: String,
+ pub evaluation_number: i32,
+ pub evaluator: Option<String>,
+ pub passed: bool,
+ pub overall_score: Option<f64>,
+ pub confidence_level: Option<String>,
+ #[sqlx(json)]
+ pub programmatic_results: serde_json::Value,
+ #[sqlx(json)]
+ pub llm_results: serde_json::Value,
+ #[sqlx(json)]
+ pub criteria_results: serde_json::Value,
+ pub summary_feedback: String,
+ pub rework_instructions: Option<String>,
+ #[sqlx(json)]
+ pub directive_snapshot: Option<serde_json::Value>,
+ #[sqlx(json)]
+ pub deliverables_snapshot: Option<serde_json::Value>,
+ pub started_at: DateTime<Utc>,
+ pub completed_at: Option<DateTime<Utc>>,
+ pub created_at: DateTime<Utc>,
+}
+
+/// Full row from directive_events table.
+#[derive(Debug, Clone, FromRow, Serialize, ToSchema)]
+#[serde(rename_all = "camelCase")]
+pub struct DirectiveEvent {
+ pub id: Uuid,
+ pub directive_id: Uuid,
+ pub chain_id: Option<Uuid>,
+ pub step_id: Option<Uuid>,
+ pub event_type: String,
+ pub severity: String,
+ #[sqlx(json)]
+ pub event_data: Option<serde_json::Value>,
+ pub actor_type: String,
+ pub actor_id: Option<Uuid>,
+ pub created_at: DateTime<Utc>,
+}
+
+/// Response for evaluation list endpoint.
+#[derive(Debug, Serialize, ToSchema)]
+#[serde(rename_all = "camelCase")]
+pub struct EvaluationListResponse {
+ pub evaluations: Vec<DirectiveEvaluation>,
+ pub total: i64,
}
diff --git a/makima/src/db/repository.rs b/makima/src/db/repository.rs
index e072eb8..d50ef61 100644
--- a/makima/src/db/repository.rs
+++ b/makima/src/db/repository.rs
@@ -11,12 +11,13 @@ use super::models::{
ContractTypeTemplateRecord, ConversationMessage, ConversationSnapshot,
CreateContractRequest, CreateDirectiveRequest, CreateFileRequest, CreateTaskRequest,
CreateTemplateRequest, Daemon, DaemonTaskAssignment, DaemonWithCapacity,
- DeliverableDefinition, Directive, DirectiveChain, DirectiveSummary,
+ DeliverableDefinition, Directive, DirectiveChain, DirectiveEvaluation, DirectiveEvent,
+ DirectiveSummary,
File, FileSummary, FileVersion, HistoryEvent, HistoryQueryFilters,
MeshChatConversation, MeshChatMessageRecord, PhaseChangeResult, PhaseConfig,
- PhaseDefinition, SupervisorHeartbeatRecord, SupervisorState, Task, TaskCheckpoint,
- TaskEvent, TaskSummary, UpdateContractRequest, UpdateDirectiveRequest,
- UpdateFileRequest, UpdateTaskRequest, UpdateTemplateRequest,
+ PhaseDefinition, StepContractSummary, SupervisorHeartbeatRecord, SupervisorState,
+ Task, TaskCheckpoint, TaskEvent, TaskSummary, UpdateContractRequest,
+ UpdateDirectiveRequest, UpdateFileRequest, UpdateTaskRequest, UpdateTemplateRequest,
};
/// Repository error types.
@@ -5184,6 +5185,29 @@ pub async fn list_steps_for_chain(
.await
}
+/// Batch-fetch lightweight contract summaries for a set of contract IDs.
+pub async fn get_contract_summaries_batch(
+ pool: &PgPool,
+ contract_ids: &[Uuid],
+) -> Result<Vec<StepContractSummary>, sqlx::Error> {
+ sqlx::query_as::<_, StepContractSummary>(
+ r#"
+ SELECT c.id, c.name, c.contract_type, c.phase, c.status,
+ COUNT(t.id) as task_count,
+ COUNT(t.id) FILTER (WHERE t.status IN ('done','merged')) as tasks_done,
+ COUNT(t.id) FILTER (WHERE t.status IN ('running','initializing','starting')) as tasks_running,
+ COUNT(t.id) FILTER (WHERE t.status = 'failed') as tasks_failed
+ FROM contracts c
+ LEFT JOIN tasks t ON t.contract_id = c.id
+ WHERE c.id = ANY($1)
+ GROUP BY c.id, c.name, c.contract_type, c.phase, c.status
+ "#,
+ )
+ .bind(contract_ids)
+ .fetch_all(pool)
+ .await
+}
+
// ── Directive orchestration functions ───────────────────────────────────────
/// Update directive status with automatic timestamp management.
@@ -5479,3 +5503,185 @@ pub async fn update_chain_status(
.fetch_optional(pool)
.await
}
+
+// ── Directive monitoring / evaluation functions ─────────────────────────────
+
+/// Create a directive evaluation record. evaluation_number is auto-incremented per step.
+pub async fn create_directive_evaluation(
+ pool: &PgPool,
+ directive_id: Uuid,
+ chain_id: Uuid,
+ step_id: Uuid,
+ contract_id: Uuid,
+ evaluation_type: &str,
+ evaluator: Option<&str>,
+ passed: bool,
+ overall_score: Option<f64>,
+ confidence_level: Option<&str>,
+ criteria_results: &serde_json::Value,
+ summary_feedback: &str,
+ rework_instructions: Option<&str>,
+) -> Result<DirectiveEvaluation, sqlx::Error> {
+ sqlx::query_as::<_, DirectiveEvaluation>(
+ r#"
+ INSERT INTO directive_evaluations (
+ directive_id, chain_id, step_id, contract_id,
+ evaluation_type, evaluation_number, evaluator,
+ passed, overall_score, confidence_level,
+ criteria_results, summary_feedback, rework_instructions,
+ completed_at
+ )
+ VALUES (
+ $1, $2, $3, $4,
+ $5, COALESCE((SELECT MAX(evaluation_number) FROM directive_evaluations WHERE step_id = $3), 0) + 1, $6,
+ $7, $8, $9,
+ $10, $11, $12,
+ NOW()
+ )
+ RETURNING *
+ "#,
+ )
+ .bind(directive_id)
+ .bind(chain_id)
+ .bind(step_id)
+ .bind(contract_id)
+ .bind(evaluation_type)
+ .bind(evaluator)
+ .bind(passed)
+ .bind(overall_score)
+ .bind(confidence_level)
+ .bind(criteria_results)
+ .bind(summary_feedback)
+ .bind(rework_instructions)
+ .fetch_one(pool)
+ .await
+}
+
+/// List evaluations for a step, ordered by evaluation_number.
+pub async fn list_evaluations_for_step(
+ pool: &PgPool,
+ step_id: Uuid,
+) -> Result<Vec<DirectiveEvaluation>, sqlx::Error> {
+ sqlx::query_as::<_, DirectiveEvaluation>(
+ r#"
+ SELECT * FROM directive_evaluations
+ WHERE step_id = $1
+ ORDER BY evaluation_number ASC
+ "#,
+ )
+ .bind(step_id)
+ .fetch_all(pool)
+ .await
+}
+
+/// Create a directive event.
+pub async fn create_directive_event(
+ pool: &PgPool,
+ directive_id: Uuid,
+ chain_id: Option<Uuid>,
+ step_id: Option<Uuid>,
+ event_type: &str,
+ severity: &str,
+ event_data: Option<&serde_json::Value>,
+ actor_type: &str,
+ actor_id: Option<Uuid>,
+) -> Result<DirectiveEvent, sqlx::Error> {
+ sqlx::query_as::<_, DirectiveEvent>(
+ r#"
+ INSERT INTO directive_events (directive_id, chain_id, step_id, event_type, severity, event_data, actor_type, actor_id)
+ VALUES ($1, $2, $3, $4, $5, $6, $7, $8)
+ RETURNING *
+ "#,
+ )
+ .bind(directive_id)
+ .bind(chain_id)
+ .bind(step_id)
+ .bind(event_type)
+ .bind(severity)
+ .bind(event_data)
+ .bind(actor_type)
+ .bind(actor_id)
+ .fetch_one(pool)
+ .await
+}
+
+/// Update step evaluation fields after an evaluation completes.
+pub async fn update_step_evaluation_fields(
+ pool: &PgPool,
+ step_id: Uuid,
+ confidence_score: Option<f64>,
+ confidence_level: Option<&str>,
+ last_evaluation_id: Uuid,
+) -> Result<Option<ChainStep>, sqlx::Error> {
+ sqlx::query_as::<_, ChainStep>(
+ r#"
+ UPDATE chain_steps
+ SET confidence_score = $2,
+ confidence_level = $3,
+ evaluation_count = evaluation_count + 1,
+ last_evaluation_id = $4
+ WHERE id = $1
+ RETURNING *
+ "#,
+ )
+ .bind(step_id)
+ .bind(confidence_score)
+ .bind(confidence_level)
+ .bind(last_evaluation_id)
+ .fetch_optional(pool)
+ .await
+}
+
+/// Update step monitoring contract/task references.
+pub async fn update_step_monitoring_contract(
+ pool: &PgPool,
+ step_id: Uuid,
+ monitoring_contract_id: Uuid,
+ monitoring_task_id: Uuid,
+) -> Result<Option<ChainStep>, sqlx::Error> {
+ sqlx::query_as::<_, ChainStep>(
+ r#"
+ UPDATE chain_steps
+ SET monitoring_contract_id = $2,
+ monitoring_task_id = $3
+ WHERE id = $1
+ RETURNING *
+ "#,
+ )
+ .bind(step_id)
+ .bind(monitoring_contract_id)
+ .bind(monitoring_task_id)
+ .fetch_optional(pool)
+ .await
+}
+
+/// Increment step rework_count.
+pub async fn increment_step_rework_count(
+ pool: &PgPool,
+ step_id: Uuid,
+) -> Result<Option<ChainStep>, sqlx::Error> {
+ sqlx::query_as::<_, ChainStep>(
+ r#"
+ UPDATE chain_steps
+ SET rework_count = rework_count + 1
+ WHERE id = $1
+ RETURNING *
+ "#,
+ )
+ .bind(step_id)
+ .fetch_optional(pool)
+ .await
+}
+
+/// Get a chain step by its monitoring contract ID.
+pub async fn get_step_by_monitoring_contract_id(
+ pool: &PgPool,
+ contract_id: Uuid,
+) -> Result<Option<ChainStep>, sqlx::Error> {
+ sqlx::query_as::<_, ChainStep>(
+ r#"SELECT * FROM chain_steps WHERE monitoring_contract_id = $1"#,
+ )
+ .bind(contract_id)
+ .fetch_optional(pool)
+ .await
+}
diff --git a/makima/src/orchestration/directive.rs b/makima/src/orchestration/directive.rs
index d17deeb..e779c18 100644
--- a/makima/src/orchestration/directive.rs
+++ b/makima/src/orchestration/directive.rs
@@ -4,8 +4,9 @@ use serde::Deserialize;
use sqlx::PgPool;
use uuid::Uuid;
+use serde::Serialize;
use crate::db::models::{
- CreateContractRequest, CreateTaskRequest, Directive, Task, UpdateContractRequest,
+ ChainStep, CreateContractRequest, CreateTaskRequest, Directive, Task, UpdateContractRequest,
};
use crate::db::repository;
use crate::server::state::SharedState;
@@ -26,6 +27,20 @@ struct ChainPlan {
steps: Vec<ChainPlanStep>,
}
+/// Result written by the monitoring supervisor after evaluating a step.
+#[derive(Debug, Deserialize, Serialize)]
+#[serde(rename_all = "camelCase")]
+struct MonitoringResult {
+ passed: bool,
+ overall_score: Option<f64>,
+ confidence_level: Option<String>,
+ #[serde(default)]
+ criteria_results: serde_json::Value,
+ #[serde(default)]
+ summary_feedback: String,
+ rework_instructions: Option<String>,
+}
+
/// Initialize a directive: create a planning contract and transition to "planning".
pub async fn init_directive(
pool: &PgPool,
@@ -195,8 +210,18 @@ pub async fn on_task_completed(
on_planning_completed(pool, state, &directive, task, owner_id).await?;
}
} else if contract.directive_id.is_some() {
- // This is a step contract completion
- on_step_completed(pool, state, &contract, task, owner_id).await?;
+ // Check if this is a monitoring contract completion
+ let monitoring_step =
+ repository::get_step_by_monitoring_contract_id(pool, contract_id)
+ .await
+ .map_err(|e| format!("Failed to check monitoring contract: {}", e))?;
+
+ if let Some(step) = monitoring_step {
+ on_monitoring_completed(pool, state, &contract, &step, task, owner_id).await?;
+ } else {
+ // This is a step contract completion
+ on_step_completed(pool, state, &contract, task, owner_id).await?;
+ }
}
Ok(())
@@ -403,32 +428,54 @@ async fn on_step_completed(
return Ok(());
};
- // Update step status based on task outcome
- let new_status = if task.status == "done" {
- "passed"
- } else {
- "failed"
- };
-
- repository::update_step_status(pool, step.id, new_status)
- .await
- .map_err(|e| format!("Failed to update step status: {}", e))?;
-
- tracing::info!(
- directive_id = %directive_id,
- step_id = %step.id,
- step_name = %step.name,
- new_status = new_status,
- "Step completed"
- );
-
- // Get the directive and advance
+ // Get the directive for threshold info
let directive = repository::get_directive(pool, directive_id)
.await
.map_err(|e| format!("Failed to get directive: {}", e))?
.ok_or("Directive not found")?;
- advance_chain(pool, state, &directive, owner_id).await
+ if task.status == "done" {
+ // Step task succeeded β€” dispatch monitoring evaluation
+ repository::update_step_status(pool, step.id, "evaluating")
+ .await
+ .map_err(|e| format!("Failed to update step status: {}", e))?;
+
+ let _ = repository::create_directive_event(
+ pool,
+ directive.id,
+ directive.current_chain_id,
+ Some(step.id),
+ "step_evaluating",
+ "info",
+ None,
+ "system",
+ None,
+ )
+ .await;
+
+ tracing::info!(
+ directive_id = %directive_id,
+ step_id = %step.id,
+ step_name = %step.name,
+ "Step task done, dispatching monitoring evaluation"
+ );
+
+ dispatch_monitoring(pool, &directive, &step, contract, owner_id).await
+ } else {
+ // Step task failed β€” mark step failed and advance
+ repository::update_step_status(pool, step.id, "failed")
+ .await
+ .map_err(|e| format!("Failed to update step status: {}", e))?;
+
+ tracing::info!(
+ directive_id = %directive_id,
+ step_id = %step.id,
+ step_name = %step.name,
+ "Step failed"
+ );
+
+ advance_chain(pool, state, &directive, owner_id).await
+ }
}
/// Check chain progress and dispatch ready steps or mark directive complete.
@@ -734,3 +781,470 @@ fn extract_plan_json(body: &[crate::db::models::BodyElement]) -> Option<String>
None
}
+
+/// Dispatch a monitoring contract to evaluate a completed step.
+async fn dispatch_monitoring(
+ pool: &PgPool,
+ directive: &Directive,
+ step: &ChainStep,
+ step_contract: &crate::db::models::Contract,
+ owner_id: Uuid,
+) -> Result<(), String> {
+ // Create monitoring contract
+ let contract = repository::create_contract_for_owner(
+ pool,
+ owner_id,
+ CreateContractRequest {
+ name: format!("{} - Monitor", step.name),
+ description: Some(format!("Monitoring evaluation for step: {}", step.name)),
+ contract_type: Some("monitoring".to_string()),
+ template_id: None,
+ initial_phase: Some("plan".to_string()),
+ autonomous_loop: Some(true),
+ phase_guard: None,
+ local_only: Some(true),
+ auto_merge_local: None,
+ },
+ )
+ .await
+ .map_err(|e| format!("Failed to create monitoring contract: {}", e))?;
+
+ // Mark contract as directive-related (not orchestrator)
+ repository::set_contract_directive_fields(pool, contract.id, Some(directive.id), false)
+ .await
+ .map_err(|e| format!("Failed to set monitoring contract directive fields: {}", e))?;
+
+ // Build evaluation prompt
+ let prompt = build_monitoring_prompt(directive, step, step_contract);
+
+ // Create supervisor task
+ let supervisor_task = repository::create_task_for_owner(
+ pool,
+ owner_id,
+ CreateTaskRequest {
+ contract_id: Some(contract.id),
+ name: format!("{} - Evaluator", step.name),
+ description: Some("Evaluate step output against directive criteria".to_string()),
+ plan: prompt,
+ parent_task_id: None,
+ is_supervisor: true,
+ priority: 8,
+ repository_url: directive.repository_url.clone(),
+ base_branch: directive.base_branch.clone(),
+ target_branch: None,
+ merge_mode: None,
+ target_repo_path: directive.local_path.clone(),
+ completion_action: None,
+ continue_from_task_id: None,
+ copy_files: None,
+ checkpoint_sha: None,
+ branched_from_task_id: None,
+ conversation_history: None,
+ supervisor_worktree_task_id: None,
+ },
+ )
+ .await
+ .map_err(|e| format!("Failed to create monitoring supervisor task: {}", e))?;
+
+ // Link supervisor to contract
+ repository::update_contract_for_owner(
+ pool,
+ contract.id,
+ owner_id,
+ UpdateContractRequest {
+ supervisor_task_id: Some(supervisor_task.id),
+ ..Default::default()
+ },
+ )
+ .await
+ .map_err(|e| match e {
+ crate::db::repository::RepositoryError::Database(e) => {
+ format!("Failed to link supervisor to monitoring contract: {}", e)
+ }
+ other => format!("Failed to link supervisor to monitoring contract: {:?}", other),
+ })?;
+
+ // Link step to monitoring contract/task
+ repository::update_step_monitoring_contract(pool, step.id, contract.id, supervisor_task.id)
+ .await
+ .map_err(|e| format!("Failed to update step monitoring contract link: {}", e))?;
+
+ // Copy repo config from directive to monitoring contract
+ if let Some(ref repo_url) = directive.repository_url {
+ let _ = repository::add_remote_repository(
+ pool,
+ contract.id,
+ "directive-repo",
+ repo_url,
+ true,
+ )
+ .await;
+ } else if let Some(ref local_path) = directive.local_path {
+ let _ = repository::add_local_repository(
+ pool,
+ contract.id,
+ "directive-repo",
+ local_path,
+ true,
+ )
+ .await;
+ }
+
+ tracing::info!(
+ directive_id = %directive.id,
+ step_id = %step.id,
+ step_name = %step.name,
+ monitoring_contract_id = %contract.id,
+ monitoring_task_id = %supervisor_task.id,
+ "Monitoring evaluation dispatched"
+ );
+
+ Ok(())
+}
+
+/// Build the monitoring supervisor prompt.
+fn build_monitoring_prompt(
+ directive: &Directive,
+ step: &ChainStep,
+ step_contract: &crate::db::models::Contract,
+) -> String {
+ format!(
+ r#"You are evaluating the output of a completed step in a directive chain.
+
+DIRECTIVE: {title}
+GOAL: {goal}
+REQUIREMENTS: {requirements}
+ACCEPTANCE CRITERIA: {acceptance_criteria}
+CONSTRAINTS: {constraints}
+
+STEP: {step_name}
+STEP DESCRIPTION: {step_description}
+STEP TASK PLAN: {task_plan}
+STEP CONTRACT ID: {step_contract_id}
+
+CONFIDENCE THRESHOLDS:
+- Green (pass): >= {threshold_green}
+- Yellow (marginal): >= {threshold_yellow}
+- Red (fail): < {threshold_yellow}
+
+Your job:
+1. Read the step contract's files to understand what was delivered:
+ makima contract files --contract-id {step_contract_id}
+ makima contract file <file_id> --contract-id {step_contract_id}
+
+2. Evaluate whether the step's output meets the directive's requirements and the step's specific task plan.
+
+3. Write your evaluation result as a JSON file named "evaluation-result" to this contract:
+ makima contract create-file "evaluation-result" < evaluation.json
+
+The JSON format:
+{{
+ "passed": true/false,
+ "overallScore": 0.0-1.0,
+ "confidenceLevel": "green" | "yellow" | "red",
+ "criteriaResults": [
+ {{
+ "criterion": "Description of what was checked",
+ "passed": true/false,
+ "score": 0.0-1.0,
+ "evidence": "Evidence supporting the assessment"
+ }}
+ ],
+ "summaryFeedback": "Brief summary of the evaluation",
+ "reworkInstructions": "If failed, specific instructions for rework (null if passed)"
+}}
+
+Scoring guidelines:
+- Score >= {threshold_green}: confidenceLevel = "green", passed = true
+- Score >= {threshold_yellow} and < {threshold_green}: confidenceLevel = "yellow", use judgment on passed
+- Score < {threshold_yellow}: confidenceLevel = "red", passed = false
+- Be specific in reworkInstructions if the step fails β€” the step will be re-executed with these instructions.
+
+After writing the evaluation file, mark the contract as complete:
+ makima supervisor complete"#,
+ title = directive.title,
+ goal = directive.goal,
+ requirements = serde_json::to_string_pretty(&directive.requirements).unwrap_or_default(),
+ acceptance_criteria = serde_json::to_string_pretty(&directive.acceptance_criteria).unwrap_or_default(),
+ constraints = serde_json::to_string_pretty(&directive.constraints).unwrap_or_default(),
+ step_name = step.name,
+ step_description = step.description.as_deref().unwrap_or("N/A"),
+ task_plan = step.task_plan.as_deref().unwrap_or("N/A"),
+ step_contract_id = step_contract.id,
+ threshold_green = directive.confidence_threshold_green,
+ threshold_yellow = directive.confidence_threshold_yellow,
+ )
+}
+
+/// Handle monitoring contract task completion β€” parse evaluation and decide step outcome.
+async fn on_monitoring_completed(
+ pool: &PgPool,
+ state: &SharedState,
+ contract: &crate::db::models::Contract,
+ step: &ChainStep,
+ task: &Task,
+ owner_id: Uuid,
+) -> Result<(), String> {
+ // Only process supervisor task completions
+ if !task.is_supervisor {
+ return Ok(());
+ }
+
+ let Some(directive_id) = contract.directive_id else {
+ return Ok(());
+ };
+
+ let directive = repository::get_directive(pool, directive_id)
+ .await
+ .map_err(|e| format!("Failed to get directive: {}", e))?
+ .ok_or("Directive not found")?;
+
+ // If monitoring task itself failed, fail-open: mark step as passed
+ if task.status == "failed" {
+ tracing::warn!(
+ directive_id = %directive_id,
+ step_id = %step.id,
+ "Monitoring task failed, fail-open: marking step as passed"
+ );
+
+ repository::update_step_status(pool, step.id, "passed")
+ .await
+ .map_err(|e| format!("Failed to update step status: {}", e))?;
+
+ let _ = repository::create_directive_event(
+ pool,
+ directive_id,
+ directive.current_chain_id,
+ Some(step.id),
+ "monitoring_failed_open",
+ "warn",
+ None,
+ "system",
+ None,
+ )
+ .await;
+
+ return advance_chain(pool, state, &directive, owner_id).await;
+ }
+
+ if task.status != "done" {
+ return Ok(());
+ }
+
+ // Read evaluation result from monitoring contract files
+ let files = repository::list_files_in_contract(pool, contract.id, owner_id)
+ .await
+ .map_err(|e| format!("Failed to list monitoring contract files: {}", e))?;
+
+ let eval_file = files.iter().find(|f| {
+ let name_lower = f.name.to_lowercase();
+ name_lower.contains("evaluation") || name_lower.contains("eval")
+ });
+
+ let eval_file = eval_file.or_else(|| files.first());
+
+ let monitoring_result = if let Some(eval_file) = eval_file {
+ let full_file = repository::get_file(pool, eval_file.id)
+ .await
+ .map_err(|e| format!("Failed to get evaluation file: {}", e))?;
+
+ if let Some(full_file) = full_file {
+ let json_str = extract_plan_json(&full_file.body);
+ json_str.and_then(|s| serde_json::from_str::<MonitoringResult>(&s).ok())
+ } else {
+ None
+ }
+ } else {
+ None
+ };
+
+ // If we couldn't parse the result, fail-open
+ let Some(result) = monitoring_result else {
+ tracing::warn!(
+ directive_id = %directive_id,
+ step_id = %step.id,
+ "Could not parse monitoring result, fail-open: marking step as passed"
+ );
+
+ repository::update_step_status(pool, step.id, "passed")
+ .await
+ .map_err(|e| format!("Failed to update step status: {}", e))?;
+
+ let _ = repository::create_directive_event(
+ pool,
+ directive_id,
+ directive.current_chain_id,
+ Some(step.id),
+ "monitoring_parse_failed_open",
+ "warn",
+ None,
+ "system",
+ None,
+ )
+ .await;
+
+ return advance_chain(pool, state, &directive, owner_id).await;
+ };
+
+ // Create evaluation record
+ let chain_id = directive.current_chain_id.unwrap_or(step.chain_id);
+ let evaluation = repository::create_directive_evaluation(
+ pool,
+ directive_id,
+ chain_id,
+ step.id,
+ contract.id,
+ "monitoring",
+ Some("automated"),
+ result.passed,
+ result.overall_score,
+ result.confidence_level.as_deref(),
+ &result.criteria_results,
+ &result.summary_feedback,
+ result.rework_instructions.as_deref(),
+ )
+ .await
+ .map_err(|e| format!("Failed to create directive evaluation: {}", e))?;
+
+ // Update step evaluation fields
+ repository::update_step_evaluation_fields(
+ pool,
+ step.id,
+ result.overall_score,
+ result.confidence_level.as_deref(),
+ evaluation.id,
+ )
+ .await
+ .map_err(|e| format!("Failed to update step evaluation fields: {}", e))?;
+
+ // Create event
+ let event_data = serde_json::json!({
+ "passed": result.passed,
+ "overallScore": result.overall_score,
+ "confidenceLevel": result.confidence_level,
+ "summaryFeedback": result.summary_feedback,
+ });
+ let _ = repository::create_directive_event(
+ pool,
+ directive_id,
+ Some(chain_id),
+ Some(step.id),
+ if result.passed { "step_evaluation_passed" } else { "step_evaluation_failed" },
+ "info",
+ Some(&event_data),
+ "system",
+ None,
+ )
+ .await;
+
+ if result.passed {
+ // Evaluation passed β€” mark step as passed
+ tracing::info!(
+ directive_id = %directive_id,
+ step_id = %step.id,
+ step_name = %step.name,
+ score = ?result.overall_score,
+ "Step evaluation passed"
+ );
+
+ repository::update_step_status(pool, step.id, "passed")
+ .await
+ .map_err(|e| format!("Failed to update step status: {}", e))?;
+
+ advance_chain(pool, state, &directive, owner_id).await
+ } else {
+ // Evaluation failed β€” check rework budget
+ let max_rework = directive.max_rework_cycles.unwrap_or(3);
+ if step.rework_count >= max_rework {
+ tracing::warn!(
+ directive_id = %directive_id,
+ step_id = %step.id,
+ step_name = %step.name,
+ rework_count = step.rework_count,
+ max_rework = max_rework,
+ "Step evaluation failed, max rework cycles exceeded"
+ );
+
+ repository::update_step_status(pool, step.id, "failed")
+ .await
+ .map_err(|e| format!("Failed to update step status: {}", e))?;
+
+ advance_chain(pool, state, &directive, owner_id).await
+ } else {
+ tracing::info!(
+ directive_id = %directive_id,
+ step_id = %step.id,
+ step_name = %step.name,
+ rework_count = step.rework_count,
+ "Step evaluation failed, scheduling rework"
+ );
+
+ repository::increment_step_rework_count(pool, step.id)
+ .await
+ .map_err(|e| format!("Failed to increment rework count: {}", e))?;
+
+ // Set step back to pending so advance_chain re-dispatches it
+ repository::update_step_status(pool, step.id, "pending")
+ .await
+ .map_err(|e| format!("Failed to update step status: {}", e))?;
+
+ advance_chain(pool, state, &directive, owner_id).await
+ }
+ }
+}
+
+/// Trigger a manual evaluation for a step. Public for use by handlers.
+pub async fn trigger_manual_evaluation(
+ pool: &PgPool,
+ _state: &SharedState,
+ owner_id: Uuid,
+ directive_id: Uuid,
+ step_id: Uuid,
+) -> Result<ChainStep, String> {
+ let directive = repository::get_directive_for_owner(pool, directive_id, owner_id)
+ .await
+ .map_err(|e| format!("Failed to get directive: {}", e))?
+ .ok_or("Directive not found")?;
+
+ // Get the step β€” find via chain steps
+ let chain_id = directive.current_chain_id.ok_or("Directive has no active chain")?;
+ let steps = repository::list_steps_for_chain(pool, chain_id)
+ .await
+ .map_err(|e| format!("Failed to list steps: {}", e))?;
+
+ let step = steps
+ .into_iter()
+ .find(|s| s.id == step_id)
+ .ok_or("Step not found in current chain")?;
+
+ // Step must have a contract_id (must have been executed)
+ let contract_id = step.contract_id.ok_or("Step has no contract β€” it hasn't been executed yet")?;
+
+ let contract = repository::get_contract_for_owner(pool, contract_id, owner_id)
+ .await
+ .map_err(|e| format!("Failed to get step contract: {}", e))?
+ .ok_or("Step contract not found")?;
+
+ // Set step to evaluating
+ let updated_step = repository::update_step_status(pool, step.id, "evaluating")
+ .await
+ .map_err(|e| format!("Failed to update step status: {}", e))?
+ .ok_or("Step not found after status update")?;
+
+ let _ = repository::create_directive_event(
+ pool,
+ directive.id,
+ directive.current_chain_id,
+ Some(step.id),
+ "manual_evaluation_triggered",
+ "info",
+ None,
+ "user",
+ None,
+ )
+ .await;
+
+ dispatch_monitoring(pool, &directive, &step, &contract, owner_id).await?;
+
+ Ok(updated_step)
+}
diff --git a/makima/src/server/handlers/directives.rs b/makima/src/server/handlers/directives.rs
index a877c6b..65f32d5 100644
--- a/makima/src/server/handlers/directives.rs
+++ b/makima/src/server/handlers/directives.rs
@@ -8,9 +8,12 @@ use axum::{
};
use uuid::Uuid;
+use std::collections::HashMap;
+
use crate::db::models::{
- ChainStep, ChainWithSteps, CreateDirectiveRequest, Directive, DirectiveChain,
- DirectiveListResponse, DirectiveWithChains, UpdateDirectiveRequest,
+ ChainStep, ChainStepWithContract, ChainWithSteps, CreateDirectiveRequest, Directive,
+ DirectiveChain, DirectiveListResponse, DirectiveWithChains, EvaluationListResponse,
+ StepContractSummary, UpdateDirectiveRequest,
};
use crate::db::repository::{self, RepositoryError};
use crate::orchestration;
@@ -123,8 +126,8 @@ pub async fn get_directive(
};
// Build chains with steps
- let mut chains_with_steps = Vec::new();
- for chain in chains {
+ let mut all_steps_by_chain = Vec::new();
+ for chain in &chains {
let steps = match repository::list_steps_for_chain(pool, chain.id).await {
Ok(s) => s,
Err(e) => {
@@ -132,11 +135,61 @@ pub async fn get_directive(
Vec::new()
}
};
- chains_with_steps.push(ChainWithSteps { chain, steps });
+ all_steps_by_chain.push(steps);
+ }
+
+ // Collect all contract IDs (from steps + orchestrator)
+ let mut contract_ids: Vec<Uuid> = all_steps_by_chain
+ .iter()
+ .flat_map(|steps| steps.iter().filter_map(|s| s.contract_id))
+ .collect();
+ if let Some(orch_id) = directive.orchestrator_contract_id {
+ contract_ids.push(orch_id);
}
+ // Batch fetch contract summaries
+ let mut summary_map: HashMap<Uuid, StepContractSummary> = if contract_ids.is_empty() {
+ HashMap::new()
+ } else {
+ match repository::get_contract_summaries_batch(pool, &contract_ids).await {
+ Ok(summaries) => summaries.into_iter().map(|s| (s.id, s)).collect(),
+ Err(e) => {
+ tracing::warn!("Failed to fetch contract summaries: {}", e);
+ HashMap::new()
+ }
+ }
+ };
+
+ // Build enriched chains
+ let chains_with_steps: Vec<ChainWithSteps> = chains
+ .into_iter()
+ .zip(all_steps_by_chain.into_iter())
+ .map(|(chain, steps)| {
+ let enriched_steps = steps
+ .into_iter()
+ .map(|step| {
+ let contract_summary =
+ step.contract_id.and_then(|id| summary_map.remove(&id));
+ ChainStepWithContract {
+ step,
+ contract_summary,
+ }
+ })
+ .collect();
+ ChainWithSteps {
+ chain,
+ steps: enriched_steps,
+ }
+ })
+ .collect();
+
+ let orchestrator_contract_summary = directive
+ .orchestrator_contract_id
+ .and_then(|id| summary_map.remove(&id));
+
Json(DirectiveWithChains {
directive,
+ orchestrator_contract_summary,
chains: chains_with_steps,
})
.into_response()
@@ -454,7 +507,37 @@ pub async fn get_chain(
}
};
- Json(ChainWithSteps { chain, steps }).into_response()
+ // Collect contract IDs from steps
+ let contract_ids: Vec<Uuid> = steps.iter().filter_map(|s| s.contract_id).collect();
+
+ let mut summary_map: HashMap<Uuid, StepContractSummary> = if contract_ids.is_empty() {
+ HashMap::new()
+ } else {
+ match repository::get_contract_summaries_batch(pool, &contract_ids).await {
+ Ok(summaries) => summaries.into_iter().map(|s| (s.id, s)).collect(),
+ Err(e) => {
+ tracing::warn!("Failed to fetch contract summaries: {}", e);
+ HashMap::new()
+ }
+ }
+ };
+
+ let enriched_steps = steps
+ .into_iter()
+ .map(|step| {
+ let contract_summary = step.contract_id.and_then(|id| summary_map.remove(&id));
+ ChainStepWithContract {
+ step,
+ contract_summary,
+ }
+ })
+ .collect();
+
+ Json(ChainWithSteps {
+ chain,
+ steps: enriched_steps,
+ })
+ .into_response()
}
/// Start a directive: create a planning contract and begin orchestration.
@@ -513,3 +596,131 @@ pub async fn start_directive(
}
}
}
+
+/// Trigger a manual evaluation for a step.
+#[utoipa::path(
+ post,
+ path = "/api/v1/directives/{id}/steps/{step_id}/evaluate",
+ params(
+ ("id" = Uuid, Path, description = "Directive ID"),
+ ("step_id" = Uuid, Path, description = "Step ID")
+ ),
+ responses(
+ (status = 200, description = "Evaluation triggered", body = ChainStep),
+ (status = 400, description = "Step cannot be evaluated", body = ApiError),
+ (status = 401, description = "Unauthorized", body = ApiError),
+ (status = 404, description = "Not found", body = ApiError),
+ (status = 503, description = "Database not configured", body = ApiError),
+ (status = 500, description = "Internal server error", body = ApiError),
+ ),
+ security(
+ ("bearer_auth" = []),
+ ("api_key" = [])
+ ),
+ tag = "Directives"
+)]
+pub async fn evaluate_step(
+ State(state): State<SharedState>,
+ Authenticated(auth): Authenticated,
+ Path((id, step_id)): Path<(Uuid, Uuid)>,
+) -> impl IntoResponse {
+ let Some(ref pool) = state.db_pool else {
+ return (
+ StatusCode::SERVICE_UNAVAILABLE,
+ Json(ApiError::new("DB_UNAVAILABLE", "Database not configured")),
+ )
+ .into_response();
+ };
+
+ match orchestration::directive::trigger_manual_evaluation(pool, &state, auth.owner_id, id, step_id).await {
+ Ok(step) => Json(step).into_response(),
+ Err(e) if e.contains("not found") || e.contains("Not found") => (
+ StatusCode::NOT_FOUND,
+ Json(ApiError::new("NOT_FOUND", e)),
+ )
+ .into_response(),
+ Err(e) if e.contains("hasn't been executed") || e.contains("no active chain") => (
+ StatusCode::BAD_REQUEST,
+ Json(ApiError::new("INVALID_STATE", e)),
+ )
+ .into_response(),
+ Err(e) => {
+ tracing::error!("Failed to trigger evaluation for step {}: {}", step_id, e);
+ (
+ StatusCode::INTERNAL_SERVER_ERROR,
+ Json(ApiError::new("EVALUATION_FAILED", e)),
+ )
+ .into_response()
+ }
+ }
+}
+
+/// List evaluations for a step.
+#[utoipa::path(
+ get,
+ path = "/api/v1/directives/{id}/steps/{step_id}/evaluations",
+ params(
+ ("id" = Uuid, Path, description = "Directive ID"),
+ ("step_id" = Uuid, Path, description = "Step ID")
+ ),
+ responses(
+ (status = 200, description = "List of evaluations", body = EvaluationListResponse),
+ (status = 401, description = "Unauthorized", body = ApiError),
+ (status = 404, description = "Not found", body = ApiError),
+ (status = 503, description = "Database not configured", body = ApiError),
+ (status = 500, description = "Internal server error", body = ApiError),
+ ),
+ security(
+ ("bearer_auth" = []),
+ ("api_key" = [])
+ ),
+ tag = "Directives"
+)]
+pub async fn list_evaluations(
+ State(state): State<SharedState>,
+ Authenticated(auth): Authenticated,
+ Path((id, step_id)): Path<(Uuid, Uuid)>,
+) -> impl IntoResponse {
+ let Some(ref pool) = state.db_pool else {
+ return (
+ StatusCode::SERVICE_UNAVAILABLE,
+ Json(ApiError::new("DB_UNAVAILABLE", "Database not configured")),
+ )
+ .into_response();
+ };
+
+ // Verify directive exists and belongs to owner
+ match repository::get_directive_for_owner(pool, id, auth.owner_id).await {
+ Ok(Some(_)) => {}
+ Ok(None) => {
+ return (
+ StatusCode::NOT_FOUND,
+ Json(ApiError::new("NOT_FOUND", "Directive not found")),
+ )
+ .into_response();
+ }
+ Err(e) => {
+ tracing::error!("Failed to get directive {}: {}", id, e);
+ return (
+ StatusCode::INTERNAL_SERVER_ERROR,
+ Json(ApiError::new("DB_ERROR", e.to_string())),
+ )
+ .into_response();
+ }
+ }
+
+ match repository::list_evaluations_for_step(pool, step_id).await {
+ Ok(evaluations) => {
+ let total = evaluations.len() as i64;
+ Json(EvaluationListResponse { evaluations, total }).into_response()
+ }
+ Err(e) => {
+ tracing::error!("Failed to list evaluations for step {}: {}", step_id, e);
+ (
+ StatusCode::INTERNAL_SERVER_ERROR,
+ Json(ApiError::new("DB_ERROR", e.to_string())),
+ )
+ .into_response()
+ }
+ }
+}
diff --git a/makima/src/server/mod.rs b/makima/src/server/mod.rs
index 1a59e12..c8242ae 100644
--- a/makima/src/server/mod.rs
+++ b/makima/src/server/mod.rs
@@ -184,6 +184,8 @@ pub fn make_router(state: SharedState) -> Router {
.route("/directives/{id}/start", post(directives::start_directive))
.route("/directives/{id}/chains", get(directives::list_chains))
.route("/directives/{id}/chains/{chain_id}", get(directives::get_chain))
+ .route("/directives/{id}/steps/{step_id}/evaluate", post(directives::evaluate_step))
+ .route("/directives/{id}/steps/{step_id}/evaluations", get(directives::list_evaluations))
// Contract supervisor resume endpoints
.route("/contracts/{id}/supervisor/resume", post(mesh_supervisor::resume_supervisor))
.route("/contracts/{id}/supervisor/conversation/rewind", post(mesh_supervisor::rewind_conversation))
diff --git a/makima/src/server/openapi.rs b/makima/src/server/openapi.rs
index 96c19e0..e680c07 100644
--- a/makima/src/server/openapi.rs
+++ b/makima/src/server/openapi.rs
@@ -4,17 +4,20 @@ use utoipa::OpenApi;
use crate::db::models::{
AddLocalRepositoryRequest, AddRemoteRepositoryRequest, BranchInfo, BranchListResponse,
- BranchTaskRequest, BranchTaskResponse, ChainStep, ChainWithSteps, ChangePhaseRequest,
+ BranchTaskRequest, BranchTaskResponse, ChainStep, ChainStepWithContract, ChainWithSteps,
+ ChangePhaseRequest,
Contract, ContractChatHistoryResponse, ContractChatMessageRecord, ContractEvent,
ContractListResponse, ContractRepository, ContractSummary, ContractWithRelations,
CreateContractRequest, CreateDirectiveRequest, CreateFileRequest,
CreateManagedRepositoryRequest, CreateTaskRequest, Daemon, DaemonDirectoriesResponse,
- DaemonDirectory, DaemonListResponse, Directive, DirectiveChain, DirectiveListResponse,
- DirectiveSummary, DirectiveWithChains, File, FileListResponse, FileSummary,
+ DaemonDirectory, DaemonListResponse, Directive, DirectiveChain, DirectiveEvaluation,
+ DirectiveEvent, DirectiveListResponse, DirectiveSummary, DirectiveWithChains,
+ EvaluationListResponse, File, FileListResponse, FileSummary,
MergeCommitRequest, MergeCompleteCheckResponse, MergeResolveRequest, MergeResultResponse,
MergeSkipRequest, MergeStartRequest, MergeStatusResponse, MeshChatConversation,
MeshChatHistoryResponse, MeshChatMessageRecord, RepositoryHistoryEntry,
- RepositoryHistoryListResponse, RepositorySuggestionsQuery, SendMessageRequest, Task,
+ RepositoryHistoryListResponse, RepositorySuggestionsQuery, SendMessageRequest,
+ StepContractSummary, Task,
TaskEventListResponse, TaskListResponse, TaskSummary, TaskWithSubtasks, TranscriptEntry,
UpdateContractRequest, UpdateDirectiveRequest, UpdateFileRequest, UpdateTaskRequest,
};
@@ -114,6 +117,8 @@ use crate::server::messages::{ApiError, AudioEncoding, StartMessage, StopMessage
directives::start_directive,
directives::list_chains,
directives::get_chain,
+ directives::evaluate_step,
+ directives::list_evaluations,
),
components(
schemas(
@@ -205,9 +210,14 @@ use crate::server::messages::{ApiError, AudioEncoding, StartMessage, StopMessage
DirectiveWithChains,
DirectiveChain,
ChainStep,
+ ChainStepWithContract,
ChainWithSteps,
+ StepContractSummary,
CreateDirectiveRequest,
UpdateDirectiveRequest,
+ DirectiveEvaluation,
+ DirectiveEvent,
+ EvaluationListResponse,
)
),
tags(
generated by cgit v1.2.3 (git 2.39.1) at 2026-05-21 10:07:25 +0000