103 lines
5.8 KiB
JavaScript
103 lines
5.8 KiB
JavaScript
import { createHash } from "node:crypto";
|
|
import { spawnSync } from "node:child_process";
|
|
import { existsSync, mkdirSync, readFileSync, writeFileSync } from "node:fs";
|
|
import { resolve } from "node:path";
|
|
|
|
const phaseIndex = process.argv.indexOf("--phase");
|
|
const phase = phaseIndex >= 0 ? process.argv[phaseIndex + 1] : "green";
|
|
if (!["red", "green"].includes(phase)) throw new Error(`Unsupported phase: ${phase}`);
|
|
const runId = process.env.DADA_TDD_RUN_ID ?? `wp3-03-${phase}-${new Date().toISOString().replace(/[^0-9]/g, "")}`;
|
|
const runDirectory = resolve("artifacts", "tdd", runId);
|
|
const casesDirectory = resolve(runDirectory, "cases");
|
|
if (existsSync(runDirectory)) throw new Error(`Evidence run already exists: ${runId}`);
|
|
mkdirSync(casesDirectory, { recursive: true });
|
|
|
|
const modelEvidence = [
|
|
"gemini-3_1-flash-image-preview",
|
|
"gemini-3-pro-image-preview",
|
|
"gpt-image-2",
|
|
].flatMap((model) => [
|
|
`${model}/contract-matrix.json`, `${model}/response.json`, `${model}/external-calls.json`, `${model}/redaction.json`,
|
|
]);
|
|
const cases = [
|
|
{
|
|
acceptance: ["AC-30", "AC-40"],
|
|
evidence: modelEvidence,
|
|
id: "TDD-WP3-ADP-001-per-model-contract",
|
|
requirements: ["GEN-01", "GEN-02", "GEN-05", "GEN-06", "GEN-13", "GEN-16"],
|
|
},
|
|
{
|
|
acceptance: ["AC-40"],
|
|
evidence: ["db-diff.json", "contract-matrix.json", "response.json"],
|
|
id: "TDD-WP3-ADP-002-invalidate-revalidate",
|
|
requirements: ["ADMIN-03", "GEN-13", "GEN-15"],
|
|
},
|
|
];
|
|
for (const item of cases) mkdirSync(resolve(casesDirectory, item.id), { recursive: true });
|
|
|
|
const externalCommands = [
|
|
["gemini-flash-external", ["validate:external", "--", "--service", "ai-gateway-service-id", "--model", "gemini-3.1-flash-image-preview", "--run-id", runId]],
|
|
["gemini-pro-external", ["validate:external", "--", "--service", "ai-gateway-service-id", "--model", "gemini-3-pro-image-preview", "--run-id", runId]],
|
|
["gpt-image-external", ["validate:external", "--", "--service", "ai-gateway-service-id", "--model", "gpt-image-2", "--run-id", runId]],
|
|
];
|
|
const commands = phase === "red"
|
|
? []
|
|
: [
|
|
["all", ["test:all"]],
|
|
["integration", ["test:integration"]],
|
|
...externalCommands,
|
|
["tdd-trace", ["validate:tdd-trace"]],
|
|
];
|
|
const environment = {
|
|
...process.env,
|
|
DADA_EVIDENCE_DIR_ADAPTERS: resolve(casesDirectory, "TDD-WP3-ADP-001-per-model-contract"),
|
|
DADA_EVIDENCE_DIR_CONTRACT: resolve(casesDirectory, "TDD-WP3-ADP-002-invalidate-revalidate"),
|
|
};
|
|
const commandResults = [];
|
|
for (const [name, args] of commands) {
|
|
const command = `pnpm ${args.join(" ")}`;
|
|
const started_at = new Date().toISOString();
|
|
const result = spawnSync(process.env.ComSpec ?? "cmd.exe", ["/d", "/s", "/c", command], { encoding: "utf8", env: environment });
|
|
if (result.stdout) process.stdout.write(result.stdout);
|
|
if (result.stderr) process.stderr.write(result.stderr);
|
|
commandResults.push({ command, exit_code: result.status ?? 1, finished_at: new Date().toISOString(), name, started_at });
|
|
}
|
|
|
|
const automatedResults = commandResults.filter((result) => !result.name.endsWith("-external"));
|
|
const automatedPassed = phase === "red"
|
|
? true
|
|
: automatedResults.every((result) => result.exit_code === 0);
|
|
const externalBlockers = phase === "green" ? ["real_gateway_credentials_absent", "external_contract_evidence_not_available"] : [];
|
|
const manifest = { path: "tasks.manifest.json", sha256: createHash("sha256").update(readFileSync("tasks.manifest.json")).digest("hex").toUpperCase() };
|
|
const commit = spawnSync("git", ["rev-parse", "--short", "HEAD"], { encoding: "utf8" }).stdout.trim();
|
|
const dirty = spawnSync("git", ["status", "--porcelain"], { encoding: "utf8" }).stdout.trim().length > 0;
|
|
const summaries = [];
|
|
for (const item of cases) {
|
|
const directory = resolve(casesDirectory, item.id);
|
|
writeFileSync(resolve(directory, "commands.json"), `${JSON.stringify({ commands: commandResults, phase, run_id: runId }, null, 2)}\n`);
|
|
const evidenceRefs = phase === "red" ? ["red-observation.json"] : item.evidence;
|
|
if (phase === "red") writeFileSync(resolve(directory, "red-observation.json"), `${JSON.stringify({
|
|
expected_failure: "Three independent provider adapters and persisted contract evidence were absent before TASK-WP3-03 implementation",
|
|
observed_commands: ["pnpm vitest run tests/integration/wp3-03-contract-evidence.test.ts"],
|
|
observed_errors: ["Cannot find module '../../apps/api/src/model-contract-evidence.js'"],
|
|
status: "red_confirmed",
|
|
}, null, 2)}\n`);
|
|
const missingEvidence = evidenceRefs.filter((file) => !existsSync(resolve(directory, file)));
|
|
const status = phase === "red"
|
|
? automatedPassed && missingEvidence.length === 0 ? "red_confirmed" : "failed"
|
|
: automatedPassed && missingEvidence.length === 0 ? "externally_blocked" : "failed";
|
|
writeFileSync(resolve(directory, "result.json"), `${JSON.stringify({
|
|
acceptance_criteria: item.acceptance, automation: ["automated"], commit, evidence_refs: evidenceRefs,
|
|
external_blockers: externalBlockers, manifest, missing_evidence: missingEvidence, phase,
|
|
requirements: item.requirements, run_id: runId, status, task_id: "TASK-WP3-03", test_id: item.id,
|
|
work_package: "WP-3", worktree_under_test: dirty ? "uncommitted implementation" : "clean committed implementation",
|
|
}, null, 2)}\n`);
|
|
summaries.push({ external_blockers: externalBlockers, missing_evidence: missingEvidence, status, test_id: item.id });
|
|
}
|
|
const status = summaries.every((item) => item.status === (phase === "red" ? "red_confirmed" : "externally_blocked"))
|
|
? phase === "red" ? "red_confirmed" : "externally_blocked"
|
|
: "failed";
|
|
writeFileSync(resolve(runDirectory, "evidence.json"), `${JSON.stringify({ cases: summaries, phase, run_id: runId, status }, null, 2)}\n`);
|
|
console.log(JSON.stringify({ cases: summaries, phase, run_id: runId, status }, null, 2));
|
|
if (status === "failed") process.exit(1);
|