ultimate-pi 0.11.0 → 0.12.0

package/.pi/harness/evals/smoke/smoke-harness-plan.mjs

@@ -0,0 +1,88 @@
+#!/usr/bin/env node
+/**
+ * smoke-harness-plan — fixture validation for plan-phase pipeline (CI).
+ * Usage: node .pi/harness/evals/smoke/smoke-harness-plan.mjs --fixture
+ */
+
+import { access, cp, mkdir, readFile, rm } from "node:fs/promises";
+import { constants } from "node:fs";
+import { dirname, join, resolve } from "node:path";
+import { fileURLToPath } from "node:url";
+import { parse as parseYaml } from "yaml";
+import { validateExecutionPlan } from "../../../scripts/validate-plan-dag.mjs";
+
+const ROOT = join(dirname(fileURLToPath(import.meta.url)), "..", "..", "..", "..");
+const FIXTURE_DIR = join(dirname(fileURLToPath(import.meta.url)), "fixtures", "plan-phase");
+
+function fail(msg) {
+	console.error(`smoke-harness-plan: FAIL: ${msg}`);
+	process.exit(1);
+}
+
+function ok(msg) {
+	console.log(`  ✓ ${msg}`);
+}
+
+async function runFixture() {
+	const fixtureRoot = join(FIXTURE_DIR, "minimal-med");
+	try {
+		await access(fixtureRoot, constants.R_OK);
+	} catch {
+		fail(`missing fixture ${fixtureRoot}`);
+	}
+
+	const packetPath = join(fixtureRoot, "plan-packet.yaml");
+	const raw = await readFile(packetPath, "utf-8");
+	const packet = parseYaml(raw);
+	if (packet.contract_version !== "1.1.0") {
+		fail("fixture contract_version must be 1.1.0");
+	}
+	if (!packet.execution_plan) fail("fixture missing execution_plan");
+
+	const { status, errors } = validateExecutionPlan(packet, fixtureRoot);
+	if (status !== "pass") {
+		fail(`DAG validation failed: ${errors.join("; ")}`);
+	}
+	ok("fixture plan-packet.yaml DAG pass");
+
+	const reviewPath = join(fixtureRoot, "plan-review.md");
+	await access(reviewPath, constants.R_OK);
+	ok("plan-review.md present");
+
+	const debateRounds = ["review-round-r1.yaml", "review-round-r4.yaml"];
+	for (const name of debateRounds) {
+		const p = join(fixtureRoot, "artifacts", name);
+		await access(p, constants.R_OK);
+		const draft = parseYaml(await readFile(p, "utf-8"));
+		if (!draft.schema_version) fail(`${name} missing schema_version`);
+	}
+	ok("debate round YAML artifacts present");
+
+	const researchPath = join(fixtureRoot, "research-brief.yaml");
+	const research = parseYaml(await readFile(researchPath, "utf-8"));
+	if (!research.decomposition || !research.hypothesis) {
+		fail("research-brief.yaml missing decomposition/hypothesis");
+	}
+	ok("research-brief.yaml structure");
+
+	console.log("smoke-harness-plan: all fixture checks passed");
+}
+
+async function main() {
+	const args = process.argv.slice(2);
+	if (args.includes("--fixture")) {
+		await runFixture();
+		return;
+	}
+	if (args.includes("--live")) {
+		console.log(
+			"smoke-harness-plan: --live requires manual /harness-plan run; skipping in CI",
+		);
+		return;
+	}
+	fail("Usage: smoke-harness-plan.mjs --fixture | --live");
+}
+
+main().catch((err) => {
+	fail(err instanceof Error ? err.message : String(err));
+});

package/.pi/harness/specs/harness-posthog-event.schema.json

@@ -24,7 +24,12 @@
 				"harness_drift_report",
 				"harness_eval_verdict",
 				"harness_sentrux_signal",
-				"harness_observation"
+				"harness_observation",
+				"harness_subagent_spawned",
+				"harness_subagent_completed",
+				"harness_subagent_result_wait",
+				"harness_subagent_setup",
+				"harness_blackboard_op"
 			]
 		},
 		"distinct_id": {

package/.pi/harness/specs/plan-execution-plan-brief.schema.json

@@ -0,0 +1,13 @@
+{
+	"$schema": "https://json-schema.org/draft/2020-12/schema",
+	"$id": "https://ultimate-pi.local/.pi/harness/specs/plan-execution-plan-brief.schema.json",
+	"title": "PlanExecutionPlanBrief",
+	"type": "object",
+	"additionalProperties": false,
+	"required": ["schema_version", "execution_plan"],
+	"properties": {
+		"schema_version": { "type": "string", "const": "1.0.0" },
+		"execution_plan": { "$ref": "plan-execution-plan.schema.json" },
+		"human_summary": { "type": "string" }
+	}
+}

package/.pi/harness/specs/plan-execution-plan.schema.json

@@ -0,0 +1,255 @@
+{
+	"$schema": "https://json-schema.org/draft/2020-12/schema",
+	"$id": "https://ultimate-pi.local/.pi/harness/specs/plan-execution-plan.schema.json",
+	"title": "ExecutionPlan",
+	"type": "object",
+	"additionalProperties": false,
+	"required": [
+		"schema_version",
+		"phases",
+		"work_items",
+		"sprint_contract",
+		"wbs_dictionary",
+		"risk_register",
+		"schedule_metadata",
+		"dag_validation"
+	],
+	"properties": {
+		"schema_version": { "type": "string", "const": "1.0.0" },
+		"phases": {
+			"type": "array",
+			"minItems": 1,
+			"items": { "$ref": "#/$defs/phase" }
+		},
+		"work_items": {
+			"type": "array",
+			"minItems": 1,
+			"items": { "$ref": "#/$defs/work_item" }
+		},
+		"sprint_contract": { "$ref": "#/$defs/sprint_contract" },
+		"wbs_dictionary": {
+			"type": "array",
+			"items": { "$ref": "#/$defs/wbs_entry" }
+		},
+		"risk_register": {
+			"type": "array",
+			"items": { "$ref": "#/$defs/risk" }
+		},
+		"schedule_metadata": { "$ref": "#/$defs/schedule_metadata" },
+		"dag_validation": { "$ref": "#/$defs/dag_validation" }
+	},
+	"$defs": {
+		"phase": {
+			"type": "object",
+			"additionalProperties": false,
+			"required": [
+				"phase_id",
+				"name",
+				"objective",
+				"entry_criteria",
+				"exit_criteria",
+				"milestone",
+				"work_item_ids"
+			],
+			"properties": {
+				"phase_id": { "type": "string", "minLength": 1 },
+				"name": { "type": "string", "minLength": 1 },
+				"objective": { "type": "string", "minLength": 1 },
+				"entry_criteria": {
+					"type": "array",
+					"minItems": 1,
+					"items": { "type": "string", "minLength": 1 }
+				},
+				"exit_criteria": {
+					"type": "array",
+					"minItems": 1,
+					"items": { "type": "string", "minLength": 1 }
+				},
+				"milestone": { "type": "string", "minLength": 1 },
+				"work_item_ids": {
+					"type": "array",
+					"minItems": 1,
+					"items": { "type": "string", "minLength": 1 }
+				}
+			}
+		},
+		"work_item": {
+			"type": "object",
+			"additionalProperties": false,
+			"required": [
+				"work_item_id",
+				"phase_id",
+				"title",
+				"description",
+				"depends_on",
+				"files",
+				"parallel_safe",
+				"done_criteria",
+				"acceptance_check_ids"
+			],
+			"properties": {
+				"work_item_id": { "type": "string", "minLength": 1 },
+				"phase_id": { "type": "string", "minLength": 1 },
+				"title": { "type": "string", "minLength": 1 },
+				"description": { "type": "string", "minLength": 1 },
+				"depends_on": {
+					"type": "array",
+					"items": { "type": "string", "minLength": 1 }
+				},
+				"files": {
+					"type": "array",
+					"items": { "type": "string", "minLength": 1 }
+				},
+				"parallel_safe": { "type": "boolean" },
+				"non_code": { "type": "boolean" },
+				"done_criteria": { "$ref": "#/$defs/done_criteria" },
+				"acceptance_check_ids": {
+					"type": "array",
+					"minItems": 1,
+					"items": { "type": "string", "minLength": 1 }
+				}
+			}
+		},
+		"done_criteria": {
+			"type": "object",
+			"additionalProperties": false,
+			"required": ["type", "spec"],
+			"properties": {
+				"type": {
+					"type": "string",
+					"enum": ["command", "test", "lint", "manual", "artifact"]
+				},
+				"spec": { "type": "string", "minLength": 1 }
+			}
+		},
+		"sprint_contract": {
+			"type": "object",
+			"additionalProperties": false,
+			"required": [
+				"in_scope",
+				"out_of_scope",
+				"definition_of_done",
+				"assumptions",
+				"external_dependencies"
+			],
+			"properties": {
+				"in_scope": {
+					"type": "array",
+					"items": { "type": "string", "minLength": 1 }
+				},
+				"out_of_scope": {
+					"type": "array",
+					"items": { "type": "string", "minLength": 1 }
+				},
+				"definition_of_done": { "type": "string", "minLength": 1 },
+				"assumptions": {
+					"type": "array",
+					"items": { "type": "string", "minLength": 1 }
+				},
+				"external_dependencies": {
+					"type": "array",
+					"items": { "type": "string", "minLength": 1 }
+				}
+			}
+		},
+		"wbs_entry": {
+			"type": "object",
+			"additionalProperties": false,
+			"required": [
+				"work_item_id",
+				"deliverable",
+				"owner_role",
+				"inputs",
+				"outputs"
+			],
+			"properties": {
+				"work_item_id": { "type": "string", "minLength": 1 },
+				"deliverable": { "type": "string", "minLength": 1 },
+				"owner_role": {
+					"type": "string",
+					"enum": ["executor", "human", "research"]
+				},
+				"inputs": {
+					"type": "array",
+					"items": { "type": "string", "minLength": 1 }
+				},
+				"outputs": {
+					"type": "array",
+					"items": { "type": "string", "minLength": 1 }
+				}
+			}
+		},
+		"risk": {
+			"type": "object",
+			"additionalProperties": false,
+			"required": [
+				"risk_id",
+				"description",
+				"likelihood",
+				"impact",
+				"mitigation",
+				"linked_work_item_ids"
+			],
+			"properties": {
+				"risk_id": { "type": "string", "minLength": 1 },
+				"description": { "type": "string", "minLength": 1 },
+				"likelihood": {
+					"type": "string",
+					"enum": ["low", "med", "high"]
+				},
+				"impact": {
+					"type": "string",
+					"enum": ["low", "med", "high"]
+				},
+				"mitigation": { "type": "string", "minLength": 1 },
+				"linked_work_item_ids": {
+					"type": "array",
+					"items": { "type": "string", "minLength": 1 }
+				}
+			}
+		},
+		"schedule_metadata": {
+			"type": "object",
+			"additionalProperties": false,
+			"required": [
+				"critical_path_work_item_ids",
+				"parallel_groups",
+				"schedule_baseline_note"
+			],
+			"properties": {
+				"critical_path_work_item_ids": {
+					"type": "array",
+					"items": { "type": "string", "minLength": 1 }
+				},
+				"parallel_groups": {
+					"type": "array",
+					"items": {
+						"type": "array",
+						"items": { "type": "string", "minLength": 1 }
+					}
+				},
+				"schedule_baseline_note": { "type": "string", "minLength": 1 }
+			}
+		},
+		"dag_validation": {
+			"type": "object",
+			"additionalProperties": false,
+			"required": ["status", "topological_order", "cycles", "conflicts"],
+			"properties": {
+				"status": { "type": "string", "enum": ["pass", "fail"] },
+				"topological_order": {
+					"type": "array",
+					"items": { "type": "string", "minLength": 1 }
+				},
+				"cycles": {
+					"type": "array",
+					"items": { "type": "array", "items": { "type": "string" } }
+				},
+				"conflicts": {
+					"type": "array",
+					"items": { "type": "string", "minLength": 1 }
+				}
+			}
+		}
+	}
+}

package/.pi/harness/specs/plan-packet.schema.json

@@ -2,7 +2,7 @@
 	"$schema": "https://json-schema.org/draft/2020-12/schema",
 	"$id": "https://ultimate-pi.local/.pi/harness/specs/plan-packet.schema.json",
 	"title": "PlanPacket",
-	"description": "Plan-gated mutation contract emitted before execution.",
+	"description": "Plan-gated mutation contract emitted before execution (instance files: plan-packet.yaml).",
 	"type": "object",
 	"additionalProperties": false,
 	"required": [
@@ -14,7 +14,8 @@
 		"assumptions",
 		"risk_level",
 		"acceptance_checks",
-		"rollback_plan"
+		"rollback_plan",
+		"execution_plan"
 	],
 	"properties": {
 		"schema_version": {
@@ -23,7 +24,7 @@
 		},
 		"contract_version": {
 			"type": "string",
-			"const": "1.0.0"
+			"enum": ["1.0.0", "1.1.0"]
 		},
 		"plan_id": {
 			"type": "string",
@@ -52,8 +53,13 @@
 			"type": "array",
 			"minItems": 1,
 			"items": {
-				"type": "string",
-				"minLength": 1
+				"type": "object",
+				"additionalProperties": false,
+				"required": ["id", "description"],
+				"properties": {
+					"id": { "type": "string", "pattern": "^AC-[0-9]+$" },
+					"description": { "type": "string", "minLength": 1 }
+				}
 			}
 		},
 		"rollback_plan": {
@@ -85,6 +91,9 @@
 					}
 				}
 			}
+		},
+		"execution_plan": {
+			"$ref": "plan-execution-plan.schema.json"
 		}
 	}
 }

package/.pi/harness/specs/plan-review-round-draft.schema.json

@@ -0,0 +1,68 @@
+{
+	"$schema": "https://json-schema.org/draft/2020-12/schema",
+	"$id": "https://ultimate-pi.local/.pi/harness/specs/plan-review-round-draft.schema.json",
+	"title": "PlanReviewRoundDraft",
+	"type": "object",
+	"additionalProperties": false,
+	"required": [
+		"schema_version",
+		"round_index",
+		"round_summary",
+		"review_gate_ready",
+		"recommended_packet_patches"
+	],
+	"properties": {
+		"schema_version": { "type": "string", "const": "1.0.0" },
+		"round_index": { "type": "integer", "minimum": 1, "maximum": 4 },
+		"debate_round_focus": {
+			"type": "string",
+			"enum": ["spec", "wbs", "schedule", "quality"]
+		},
+		"round_summary": { "type": "string", "minLength": 1 },
+		"validation_summary": { "type": "string" },
+		"adversary_summary": { "type": "string" },
+		"disputes": {
+			"type": "array",
+			"items": { "type": "string", "minLength": 1 }
+		},
+		"recommended_packet_patches": {
+			"type": "array",
+			"items": {
+				"type": "object",
+				"required": ["path", "value"],
+				"properties": {
+					"path": { "type": "string", "minLength": 1 },
+					"value": {}
+				}
+			}
+		},
+		"review_gate_ready": { "type": "boolean" },
+		"participants": {
+			"type": "array",
+			"items": { "type": "string", "minLength": 1 }
+		},
+		"claims": { "type": "array", "items": { "type": "string" } },
+		"rebuttals": { "type": "array", "items": { "type": "string" } },
+		"evidence_refs": { "type": "array", "items": { "type": "string" } },
+		"token_usage": {
+			"type": "object",
+			"properties": {
+				"per_agent": {
+					"type": "object",
+					"additionalProperties": { "type": "number" }
+				},
+				"round_total": { "type": "number" }
+			}
+		},
+		"consensus_delta": { "type": "number" },
+		"severity_scores": {
+			"type": "object",
+			"properties": {
+				"correctness": { "type": "number" },
+				"security": { "type": "number" },
+				"architecture": { "type": "number" },
+				"test_integrity": { "type": "number" }
+			}
+		}
+	}
+}

package/.pi/harness/specs/plan-sprint-audit-turn.schema.json

@@ -0,0 +1,29 @@
+{
+	"$schema": "https://json-schema.org/draft/2020-12/schema",
+	"$id": "https://ultimate-pi.local/.pi/harness/specs/plan-sprint-audit-turn.schema.json",
+	"title": "PlanSprintAuditTurn",
+	"type": "object",
+	"additionalProperties": false,
+	"required": ["schema_version", "round_index", "gaps", "recommendation"],
+	"properties": {
+		"schema_version": { "type": "string", "const": "1.0.0" },
+		"round_index": { "type": "integer", "minimum": 1, "maximum": 4 },
+		"gaps": {
+			"type": "array",
+			"items": { "type": "string", "minLength": 1 }
+		},
+		"checkpoint_issues": {
+			"type": "array",
+			"items": { "type": "string", "minLength": 1 }
+		},
+		"done_criteria_issues": {
+			"type": "array",
+			"items": { "type": "string", "minLength": 1 }
+		},
+		"recommendation": {
+			"type": "string",
+			"enum": ["proceed", "revise"]
+		},
+		"human_summary": { "type": "string" }
+	}
+}

package/.pi/harness/specs/plan-stack-brief.schema.json

@@ -0,0 +1,65 @@
+{
+	"$schema": "https://json-schema.org/draft/2020-12/schema",
+	"$id": "https://ultimate-pi.local/.pi/harness/specs/plan-stack-brief.schema.json",
+	"title": "PlanStackBrief",
+	"type": "object",
+	"additionalProperties": false,
+	"required": [
+		"schema_version",
+		"problem_framing",
+		"constraints",
+		"options",
+		"recommended_primary",
+		"rationale"
+	],
+	"properties": {
+		"schema_version": { "type": "string", "const": "1.0.0" },
+		"problem_framing": { "type": "string", "minLength": 1 },
+		"constraints": {
+			"type": "array",
+			"items": { "type": "string", "minLength": 1 }
+		},
+		"options": {
+			"type": "array",
+			"minItems": 1,
+			"items": { "$ref": "#/$defs/stack_option" }
+		},
+		"recommended_primary": { "type": "string", "minLength": 1 },
+		"rationale": { "type": "string", "minLength": 1 },
+		"open_questions": {
+			"type": "array",
+			"items": { "type": "string", "minLength": 1 }
+		}
+	},
+	"$defs": {
+		"stack_option": {
+			"type": "object",
+			"additionalProperties": false,
+			"required": [
+				"name",
+				"category",
+				"fit_summary",
+				"tradeoffs",
+				"risks",
+				"evidence_refs",
+				"recommendation_rank"
+			],
+			"properties": {
+				"name": { "type": "string" },
+				"category": { "type": "string" },
+				"fit_summary": { "type": "string" },
+				"tradeoffs": {
+					"type": "object",
+					"required": ["pros", "cons"],
+					"properties": {
+						"pros": { "type": "array", "items": { "type": "string" } },
+						"cons": { "type": "array", "items": { "type": "string" } }
+					}
+				},
+				"risks": { "type": "array", "items": { "type": "string" } },
+				"evidence_refs": { "type": "array", "items": { "type": "string" } },
+				"recommendation_rank": { "type": "integer", "minimum": 1 }
+			}
+		}
+	}
+}

package/.pi/harness/specs/plan-validation-turn.schema.json

@@ -0,0 +1,42 @@
+{
+	"$schema": "https://json-schema.org/draft/2020-12/schema",
+	"$id": "https://ultimate-pi.local/.pi/harness/specs/plan-validation-turn.schema.json",
+	"title": "PlanValidationTurn",
+	"type": "object",
+	"additionalProperties": false,
+	"required": [
+		"schema_version",
+		"round_index",
+		"debate_round_focus",
+		"checks",
+		"overall_ready"
+	],
+	"properties": {
+		"schema_version": { "type": "string", "const": "1.0.0" },
+		"round_index": { "type": "integer", "minimum": 1, "maximum": 4 },
+		"debate_round_focus": {
+			"type": "string",
+			"enum": ["spec", "wbs", "schedule", "quality"]
+		},
+		"checks": {
+			"type": "array",
+			"items": {
+				"type": "object",
+				"required": ["id", "status", "evidence"],
+				"properties": {
+					"id": { "type": "string" },
+					"status": { "type": "string", "enum": ["pass", "warn", "fail"] },
+					"evidence": { "type": "string" },
+					"work_item_id": { "type": "string" },
+					"phase_id": { "type": "string" }
+				}
+			}
+		},
+		"mitigations": {
+			"type": "array",
+			"items": { "type": "string", "minLength": 1 }
+		},
+		"overall_ready": { "type": "boolean" },
+		"human_summary": { "type": "string" }
+	}
+}

package/.pi/harness/specs/round-result.schema.json

@@ -46,7 +46,17 @@
 			"minItems": 2,
 			"items": {
 				"type": "string",
-				"enum": ["EvaluatorAgent", "AdversaryAgent", "TieBreakerAgent"]
+				"enum": [
+					"EvaluatorAgent",
+					"AdversaryAgent",
+					"TieBreakerAgent",
+					"PlanEvaluatorAgent",
+					"PlanAdversaryAgent",
+					"HypothesisValidatorAgent",
+					"SprintContractAuditorAgent",
+					"ReviewIntegratorAgent",
+					"StackResearchAgent"
+				]
 			}
 		},
 		"claims": {
@@ -80,7 +90,7 @@
 					"additionalProperties": {
 						"type": "integer",
 						"minimum": 0,
-						"maximum": 2500
+						"maximum": 35000
 					}
 				},
 				"round_total": {
@@ -101,19 +111,16 @@
 			"properties": {
 				"name": {
 					"type": "string",
-					"const": "aggressive"
+					"enum": ["aggressive", "plan"]
 				},
 				"max_rounds": {
-					"type": "integer",
-					"const": 6
+					"type": "integer"
 				},
 				"round_token_cap": {
-					"type": "integer",
-					"const": 2500
+					"type": "integer"
 				},
 				"debate_global_cap": {
-					"type": "integer",
-					"const": 35000
+					"type": "integer"
 				}
 			}
 		},