ultimate-pi 0.2.5 → 0.2.6

package/.pi/harness/router/propose-router-tuning.mjs

@@ -1,149 +0,0 @@
-#!/usr/bin/env node
-
-import crypto from "node:crypto";
-import fs from "node:fs";
-import path from "node:path";
-
-const ROUTER_PATH = ".pi/model-router.json";
-
-function fail(message) {
-	process.stderr.write(`Error: ${message}\n`);
-	process.exit(1);
-}
-
-function parseArgs(argv) {
-	const args = {};
-	for (let i = 0; i < argv.length; i++) {
-		const token = argv[i];
-		if (!token.startsWith("--")) continue;
-		const key = token.slice(2);
-		const value = argv[i + 1];
-		if (!value || value.startsWith("--")) {
-			args[key] = true;
-			continue;
-		}
-		args[key] = value;
-		i++;
-	}
-	return args;
-}
-
-function readJson(filePath, label) {
-	if (!fs.existsSync(filePath)) {
-		fail(`${label} not found: ${filePath}`);
-	}
-	try {
-		return JSON.parse(fs.readFileSync(filePath, "utf8"));
-	} catch (error) {
-		fail(`${label} is not valid JSON (${filePath}): ${error.message}`);
-	}
-}
-
-function sha256FromJson(value) {
-	const canonical = `${JSON.stringify(value, null, 2)}\n`;
-	return crypto.createHash("sha256").update(canonical).digest("hex");
-}
-
-function ensureEvidence(evidence) {
-	const required = [
-		"sample_count",
-		"min_sample_count",
-		"success_rate_delta",
-		"cost_per_task_delta",
-		"regression_guard_passed",
-		"trace_refs",
-	];
-	for (const field of required) {
-		if (!(field in evidence)) fail(`evidence missing required field: ${field}`);
-	}
-	if (!Number.isInteger(evidence.sample_count) || evidence.sample_count < 1) {
-		fail("evidence.sample_count must be an integer >= 1");
-	}
-	if (
-		!Number.isInteger(evidence.min_sample_count) ||
-		evidence.min_sample_count < 1
-	) {
-		fail("evidence.min_sample_count must be an integer >= 1");
-	}
-	if (evidence.sample_count < evidence.min_sample_count) {
-		fail(
-			`insufficient sample_count (${evidence.sample_count} < ${evidence.min_sample_count})`,
-		);
-	}
-	if (typeof evidence.success_rate_delta !== "number") {
-		fail("evidence.success_rate_delta must be numeric");
-	}
-	if (typeof evidence.cost_per_task_delta !== "number") {
-		fail("evidence.cost_per_task_delta must be numeric");
-	}
-	if (evidence.regression_guard_passed !== true) {
-		fail("evidence.regression_guard_passed must be true");
-	}
-	if (!Array.isArray(evidence.trace_refs) || evidence.trace_refs.length === 0) {
-		fail("evidence.trace_refs must be a non-empty array");
-	}
-}
-
-const args = parseArgs(process.argv.slice(2));
-
-if (args.help || args.h) {
-	process.stdout.write(
-		[
-			"Usage:",
-			"  node .pi/harness/router/propose-router-tuning.mjs \\",
-			"    --evidence <evidence.json> \\",
-			"    --candidate <candidate-router.json> \\",
-			"    --proposal-out <proposal.json>",
-			"",
-			"Behavior:",
-			"  - validates evidence thresholds",
-			"  - captures base/candidate router hashes",
-			"  - emits proposal artifact without changing .pi/model-router.json",
-		].join("\n"),
-	);
-	process.exit(0);
-}
-
-if (!args.evidence) fail("missing --evidence");
-if (!args.candidate) fail("missing --candidate");
-if (!args["proposal-out"]) fail("missing --proposal-out");
-
-const baseRouter = readJson(ROUTER_PATH, "base router");
-const candidateRouter = readJson(args.candidate, "candidate router");
-const evidence = readJson(args.evidence, "evidence");
-
-ensureEvidence(evidence);
-
-const now = new Date().toISOString();
-const proposalId = `router-tune-${now.replace(/[:.]/g, "-")}`;
-
-const proposal = {
-	schema_version: "1.0.0",
-	proposal_id: proposalId,
-	created_at: now,
-	router_path: ROUTER_PATH,
-	base_router_sha256: sha256FromJson(baseRouter),
-	candidate_router_sha256: sha256FromJson(candidateRouter),
-	evidence,
-	status: "proposed",
-	approval: {
-		required: true,
-		approved_by: null,
-		approved_at: null,
-		justification: null,
-	},
-	candidate_router: candidateRouter,
-};
-
-const outputPath = path.resolve(args["proposal-out"]);
-fs.mkdirSync(path.dirname(outputPath), { recursive: true });
-fs.writeFileSync(outputPath, `${JSON.stringify(proposal, null, 2)}\n`);
-
-process.stdout.write(
-	[
-		"Router tuning proposal created.",
-		`proposal_id: ${proposal.proposal_id}`,
-		`output: ${outputPath}`,
-		"status: proposed (no router write performed)",
-	].join("\n") + "\n",
-);

package/.pi/harness/runs/019e272f-3eef-7107-9712-ce281de55707-1778773891854/events.jsonl

@@ -1,2 +0,0 @@
-{"timestamp":"2026-05-14T15:51:31.965Z","type":"run_start","run_id":"019e272f-3eef-7107-9712-ce281de55707-1778773891854","plan_id":"plan-unknown","phase":"plan"}
-{"timestamp":"2026-05-14T15:51:38.346Z","type":"run_end","run_id":"019e272f-3eef-7107-9712-ce281de55707-1778773891854","phase":"plan","tool_span_count":0,"artifact_ref_count":0}

package/.pi/harness/runs/019e272f-3eef-7107-9712-ce281de55707-1778773891854/trace.json

@@ -1,17 +0,0 @@
-{
-	"schema_version": "1.0.0",
-	"contract_version": "1.0.0",
-	"run_id": "019e272f-3eef-7107-9712-ce281de55707-1778773891854",
-	"plan_id": "plan-unknown",
-	"agent_id": "019e272f-3eef-7107-9712-ce281de55707",
-	"phase": "plan",
-	"model": "auto",
-	"thinking_level": "off",
-	"tool_spans": [],
-	"artifact_refs": [],
-	"cost": {
-		"input_tokens": 15381,
-		"output_tokens": 33,
-		"total_tokens": 15414
-	}
-}

package/.pi/harness/runs/019e272f-3eef-7107-9712-ce281de55707-1778773912057/events.jsonl

@@ -1,2 +0,0 @@
-{"timestamp":"2026-05-14T15:51:52.062Z","type":"run_start","run_id":"019e272f-3eef-7107-9712-ce281de55707-1778773912057","plan_id":"plan-unknown","phase":"plan"}
-{"timestamp":"2026-05-14T15:52:14.313Z","type":"run_end","run_id":"019e272f-3eef-7107-9712-ce281de55707-1778773912057","phase":"plan","tool_span_count":0,"artifact_ref_count":0}

package/.pi/harness/runs/019e272f-3eef-7107-9712-ce281de55707-1778773912057/trace.json

@@ -1,17 +0,0 @@
-{
-	"schema_version": "1.0.0",
-	"contract_version": "1.0.0",
-	"run_id": "019e272f-3eef-7107-9712-ce281de55707-1778773912057",
-	"plan_id": "plan-unknown",
-	"agent_id": "019e272f-3eef-7107-9712-ce281de55707",
-	"phase": "plan",
-	"model": "auto",
-	"thinking_level": "off",
-	"tool_spans": [],
-	"artifact_refs": [],
-	"cost": {
-		"input_tokens": 31337,
-		"output_tokens": 528,
-		"total_tokens": 31865
-	}
-}

package/.pi/harness/runs/019e2732-8651-74e5-9f5d-4d06c3105f25-1778774086096/events.jsonl

@@ -1,6 +0,0 @@
-{"timestamp":"2026-05-14T15:54:46.136Z","type":"run_start","run_id":"019e2732-8651-74e5-9f5d-4d06c3105f25-1778774086096","plan_id":"plan-unknown","phase":"plan"}
-{"timestamp":"2026-05-14T15:54:59.110Z","type":"tool_start","run_id":"019e2732-8651-74e5-9f5d-4d06c3105f25-1778774086096","tool_call_id":"call_00_7UHDcydTHJHVR2dT5xpb0903","tool_name":"bash"}
-{"timestamp":"2026-05-14T15:54:59.137Z","type":"tool_start","run_id":"019e2732-8651-74e5-9f5d-4d06c3105f25-1778774086096","tool_call_id":"call_01_aNsry1whTl5hRf5Ew91t3142","tool_name":"bash"}
-{"timestamp":"2026-05-14T15:54:59.139Z","type":"tool_start","run_id":"019e2732-8651-74e5-9f5d-4d06c3105f25-1778774086096","tool_call_id":"call_02_N2e56Q6vKr6cAYzd4Z9q7953","tool_name":"bash"}
-{"timestamp":"2026-05-14T15:55:11.546Z","type":"tool_start","run_id":"019e2732-8651-74e5-9f5d-4d06c3105f25-1778774086096","tool_call_id":"call_00_wG71Rv3SKrf6R9K03EeS0264","tool_name":"ctx_batch_execute"}
-{"timestamp":"2026-05-14T15:55:25.167Z","type":"run_end","run_id":"019e2732-8651-74e5-9f5d-4d06c3105f25-1778774086096","phase":"plan","tool_span_count":4,"artifact_ref_count":0}

package/.pi/harness/runs/019e2732-8651-74e5-9f5d-4d06c3105f25-1778774086096/trace.json

@@ -1,42 +0,0 @@
-{
-	"schema_version": "1.0.0",
-	"contract_version": "1.0.0",
-	"run_id": "019e2732-8651-74e5-9f5d-4d06c3105f25-1778774086096",
-	"plan_id": "plan-unknown",
-	"agent_id": "019e2732-8651-74e5-9f5d-4d06c3105f25",
-	"phase": "plan",
-	"model": "auto",
-	"thinking_level": "off",
-	"tool_spans": [
-		{
-			"tool_call_id": "call_00_7UHDcydTHJHVR2dT5xpb0903",
-			"tool_name": "bash",
-			"started_at": "2026-05-14T15:54:59.108Z",
-			"ended_at": "2026-05-14T15:54:59.108Z"
-		},
-		{
-			"tool_call_id": "call_01_aNsry1whTl5hRf5Ew91t3142",
-			"tool_name": "bash",
-			"started_at": "2026-05-14T15:54:59.136Z",
-			"ended_at": "2026-05-14T15:54:59.136Z"
-		},
-		{
-			"tool_call_id": "call_02_N2e56Q6vKr6cAYzd4Z9q7953",
-			"tool_name": "bash",
-			"started_at": "2026-05-14T15:54:59.139Z",
-			"ended_at": "2026-05-14T15:54:59.139Z"
-		},
-		{
-			"tool_call_id": "call_00_wG71Rv3SKrf6R9K03EeS0264",
-			"tool_name": "ctx_batch_execute",
-			"started_at": "2026-05-14T15:55:11.541Z",
-			"ended_at": "2026-05-14T15:55:11.541Z"
-		}
-	],
-	"artifact_refs": [],
-	"cost": {
-		"input_tokens": 16951,
-		"output_tokens": 1020,
-		"total_tokens": 17971
-	}
-}

package/.pi/harness/runs/019e2732-8651-74e5-9f5d-4d06c3105f25-1778774136101/events.jsonl

@@ -1 +0,0 @@
-{"timestamp":"2026-05-14T15:55:36.107Z","type":"run_start","run_id":"019e2732-8651-74e5-9f5d-4d06c3105f25-1778774136101","plan_id":"plan-unknown","phase":"plan"}

package/.pi/harness/runs/019e2758-b332-771b-ad6f-54d0d8478768-1778776600591/events.jsonl

@@ -1,2 +0,0 @@
-{"timestamp":"2026-05-14T16:36:40.660Z","type":"run_start","run_id":"019e2758-b332-771b-ad6f-54d0d8478768-1778776600591","plan_id":"plan-unknown","phase":"plan"}
-{"timestamp":"2026-05-14T16:36:47.570Z","type":"run_end","run_id":"019e2758-b332-771b-ad6f-54d0d8478768-1778776600591","phase":"plan","tool_span_count":0,"artifact_ref_count":0}

package/.pi/harness/runs/019e2758-b332-771b-ad6f-54d0d8478768-1778776600591/trace.json

@@ -1,17 +0,0 @@
-{
-	"schema_version": "1.0.0",
-	"contract_version": "1.0.0",
-	"run_id": "019e2758-b332-771b-ad6f-54d0d8478768-1778776600591",
-	"plan_id": "plan-unknown",
-	"agent_id": "019e2758-b332-771b-ad6f-54d0d8478768",
-	"phase": "plan",
-	"model": "auto",
-	"thinking_level": "off",
-	"tool_spans": [],
-	"artifact_refs": [],
-	"cost": {
-		"input_tokens": 21,
-		"output_tokens": 32,
-		"total_tokens": 53
-	}
-}

package/.pi/harness/runs/README.md

@@ -1,6 +0,0 @@
-# Harness Runs
-
-Store lightweight run metadata and trace indexes here.
-
-- Primary source of truth for full trace payloads remains external telemetry.
-- Local files should contain run IDs, pointers, and replay metadata only.

package/.pi/harness/runs/budget-events.jsonl

@@ -1,4 +0,0 @@
-{"timestamp":"2026-05-14T15:54:59.134Z","schema_version":"1.0.0","contract_version":"1.0.0","event_type":"budget_exhausted","run_id":"019e2732-8651-74e5-9f5d-4d06c3105f25","debate_id":"plan-budget-guard","round_count":1,"budget_used":16593,"exhaustion_reason":"debate_global_cap_exceeded","caps":{"max_rounds":6,"round_token_cap":2500,"debate_global_cap":35000},"minimum_evidence_confidence":0.6,"default_policy_outcome":"block","human_override_allowed":true}
-{"timestamp":"2026-05-14T15:54:59.138Z","schema_version":"1.0.0","contract_version":"1.0.0","event_type":"budget_exhausted","run_id":"019e2732-8651-74e5-9f5d-4d06c3105f25","debate_id":"plan-budget-guard","round_count":1,"budget_used":16593,"exhaustion_reason":"debate_global_cap_exceeded","caps":{"max_rounds":6,"round_token_cap":2500,"debate_global_cap":35000},"minimum_evidence_confidence":0.6,"default_policy_outcome":"block","human_override_allowed":true}
-{"timestamp":"2026-05-14T15:54:59.140Z","schema_version":"1.0.0","contract_version":"1.0.0","event_type":"budget_exhausted","run_id":"019e2732-8651-74e5-9f5d-4d06c3105f25","debate_id":"plan-budget-guard","round_count":1,"budget_used":16593,"exhaustion_reason":"debate_global_cap_exceeded","caps":{"max_rounds":6,"round_token_cap":2500,"debate_global_cap":35000},"minimum_evidence_confidence":0.6,"default_policy_outcome":"block","human_override_allowed":true}
-{"timestamp":"2026-05-14T15:55:11.581Z","schema_version":"1.0.0","contract_version":"1.0.0","event_type":"budget_exhausted","run_id":"019e2732-8651-74e5-9f5d-4d06c3105f25","debate_id":"plan-budget-guard","round_count":1,"budget_used":17161,"exhaustion_reason":"debate_global_cap_exceeded","caps":{"max_rounds":6,"round_token_cap":2500,"debate_global_cap":35000},"minimum_evidence_confidence":0.6,"default_policy_outcome":"block","human_override_allowed":true}

package/.pi/harness/runs/canary-candidate-router.json

@@ -1,72 +0,0 @@
-{
-	"defaultProfile": "auto",
-	"debug": false,
-	"classifierModel": "opencode-go/qwen3.6-plus",
-	"phaseBias": 0.5,
-	"maxSessionBudget": 1.0,
-	"largeContextThreshold": 100000,
-	"rules": [
-		{
-			"matches": ["deploy", "production", "release"],
-			"tier": "high",
-			"reason": "Safety check for production tasks"
-		},
-		{
-			"matches": "changelog",
-			"tier": "low"
-		}
-	],
-	"profiles": {
-		"auto": {
-			"high": {
-				"model": "opencode-go/deepseek-v4-pro",
-				"thinking": "high",
-				"fallbacks": ["opencode-go/qwen3.6-plus", "opencode-go/kimi-k2.6"]
-			},
-			"medium": {
-				"model": "opencode-go/qwen3.6-plus",
-				"thinking": "medium",
-				"fallbacks": ["opencode-go/deepseek-v4-pro"]
-			},
-			"low": {
-				"model": "opencode-go/deepseek-v4-flash",
-				"thinking": "low",
-				"fallbacks": ["opencode-go/qwen3.5-plus"]
-			}
-		},
-		"cheap": {
-			"high": {
-				"model": "opencode-go/qwen3.6-plus",
-				"thinking": "low",
-				"fallbacks": ["opencode-go/qwen3.5-plus"]
-			},
-			"medium": {
-				"model": "opencode-go/qwen3.5-plus",
-				"thinking": "off",
-				"fallbacks": ["opencode-go/deepseek-v4-flash"]
-			},
-			"low": {
-				"model": "opencode-go/deepseek-v4-flash",
-				"thinking": "off",
-				"fallbacks": ["opencode-go/qwen3.5-plus"]
-			}
-		},
-		"deep": {
-			"high": {
-				"model": "opencode-go/deepseek-v4-pro",
-				"thinking": "xhigh",
-				"fallbacks": ["opencode-go/kimi-k2.6"]
-			},
-			"medium": {
-				"model": "opencode-go/kimi-k2.6",
-				"thinking": "medium",
-				"fallbacks": ["opencode-go/deepseek-v4-pro"]
-			},
-			"low": {
-				"model": "opencode-go/qwen3.6-plus",
-				"thinking": "low",
-				"fallbacks": ["opencode-go/deepseek-v4-flash"]
-			}
-		}
-	}
-}

package/.pi/harness/runs/canary-evidence.json

@@ -1,9 +0,0 @@
-{
-	"sample_count": 24,
-	"min_sample_count": 12,
-	"success_rate_delta": 0.08,
-	"cost_per_task_delta": -0.04,
-	"regression_guard_passed": true,
-	"trace_refs": ["run-canary-001", "run-canary-002"],
-	"notes": "canary validation synthetic evidence"
-}

package/.pi/harness/runs/index.jsonl

@@ -1,4 +0,0 @@
-{"timestamp":"2026-05-14T15:51:38.345Z","run_id":"019e272f-3eef-7107-9712-ce281de55707-1778773891854","plan_id":"plan-unknown","phase":"plan","trace_file":"/home/aryaniyaps/ai-projects/ultimate-pi/.pi/harness/runs/019e272f-3eef-7107-9712-ce281de55707-1778773891854/trace.json"}
-{"timestamp":"2026-05-14T15:52:14.312Z","run_id":"019e272f-3eef-7107-9712-ce281de55707-1778773912057","plan_id":"plan-unknown","phase":"plan","trace_file":"/home/aryaniyaps/ai-projects/ultimate-pi/.pi/harness/runs/019e272f-3eef-7107-9712-ce281de55707-1778773912057/trace.json"}
-{"timestamp":"2026-05-14T15:55:25.166Z","run_id":"019e2732-8651-74e5-9f5d-4d06c3105f25-1778774086096","plan_id":"plan-unknown","phase":"plan","trace_file":"/home/aryaniyaps/ai-projects/ultimate-pi/.pi/harness/runs/019e2732-8651-74e5-9f5d-4d06c3105f25-1778774086096/trace.json"}
-{"timestamp":"2026-05-14T16:36:47.569Z","run_id":"019e2758-b332-771b-ad6f-54d0d8478768-1778776600591","plan_id":"plan-unknown","phase":"plan","trace_file":"/home/aryaniyaps/ai-projects/ultimate-pi/.pi/harness/runs/019e2758-b332-771b-ad6f-54d0d8478768-1778776600591/trace.json"}

package/.pi/model-router.json

@@ -1,95 +0,0 @@
-{
-    "defaultProfile": "auto",
-    "debug": false,
-    "classifierModel": "opencode-go/qwen3.6-plus",
-    "phaseBias": 0.5,
-    "maxSessionBudget": 1.0,
-    "largeContextThreshold": 100000,
-    "rules": [
-        {
-            "matches": [
-                "deploy",
-                "production",
-                "release"
-            ],
-            "tier": "high",
-            "reason": "Safety check for production tasks"
-        },
-        {
-            "matches": "changelog",
-            "tier": "low"
-        }
-    ],
-    "profiles": {
-        "auto": {
-            "high": {
-                "model": "opencode-go/deepseek-v4-pro",
-                "thinking": "high",
-                "fallbacks": [
-                    "opencode-go/qwen3.6-plus",
-                    "opencode-go/kimi-k2.6"
-                ]
-            },
-            "medium": {
-                "model": "opencode-go/qwen3.6-plus",
-                "thinking": "medium",
-                "fallbacks": [
-                    "opencode-go/deepseek-v4-pro"
-                ]
-            },
-            "low": {
-                "model": "opencode-go/deepseek-v4-flash",
-                "thinking": "low",
-                "fallbacks": [
-                    "opencode-go/qwen3.5-plus"
-                ]
-            }
-        },
-        "cheap": {
-            "high": {
-                "model": "opencode-go/qwen3.6-plus",
-                "thinking": "low",
-                "fallbacks": [
-                    "opencode-go/qwen3.5-plus"
-                ]
-            },
-            "medium": {
-                "model": "opencode-go/qwen3.5-plus",
-                "thinking": "off",
-                "fallbacks": [
-                    "opencode-go/deepseek-v4-flash"
-                ]
-            },
-            "low": {
-                "model": "opencode-go/deepseek-v4-flash",
-                "thinking": "off",
-                "fallbacks": [
-                    "opencode-go/qwen3.5-plus"
-                ]
-            }
-        },
-        "deep": {
-            "high": {
-                "model": "opencode-go/deepseek-v4-pro",
-                "thinking": "xhigh",
-                "fallbacks": [
-                    "opencode-go/kimi-k2.6"
-                ]
-            },
-            "medium": {
-                "model": "opencode-go/kimi-k2.6",
-                "thinking": "medium",
-                "fallbacks": [
-                    "opencode-go/deepseek-v4-pro"
-                ]
-            },
-            "low": {
-                "model": "opencode-go/qwen3.6-plus",
-                "thinking": "low",
-                "fallbacks": [
-                    "opencode-go/deepseek-v4-flash"
-                ]
-            }
-        }
-    }
-}

package/.pi/npm/.gitignore

@@ -1,2 +0,0 @@
-*
-!.gitignore