[Feature] Centralize config: env asserts, magic numbers, fetch timeouts, settings re-read

[chronoai-shining]

Severity: Medium

Configuration is scattered across process.env[...] reads in every module, with magic numbers inline, no required-env assertions, and Settings stored in Mongo but never re-read after boot.

Problems

1. No required-env assertions at boot.

• sisyphus-api/src/db.ts:6-7, runner/aevatar-client.ts:5-7, runner/session-manager.ts:21-22, ingest/chrono-graph-client.ts:5-6, workflows/ornn-client.ts:5, workflows/mainnet-client.ts:6-7, workflows/reference-resolver.ts:6-7, papers/storage-client.ts:5-6 — all use process.env["X"] ?? "http://localhost:NNNN" fallbacks. In a k8s pod, the localhost fallback hits the pod itself and produces opaque ECONNREFUSED instead of "config missing".
• GRAPH_ID defaults to empty string. The URL ${CHRONO_GRAPH_URL}/api/graphs//nodes (note empty segment) gets fired against the upstream silently.

2. Magic numbers everywhere.

• Timeout literals 5000 / 10000 / 15000 / 30000 / 120000 ms scattered across aevatar-client.ts, storage-client.ts, chrono-graph-client.ts, mainnet-client.ts, ornn-client.ts.
• MAX_NODES = 50_000 (papers/controllers/CompileController.ts:61).
• MAX_CONCURRENT = 3, COMPILE_TIMEOUT_MS = 300_000 (papers/compiler.ts:9-10).
• MAX_CACHE_SIZE = 500 (schemas/validator.ts:8).
• TTLs 365 * 24 * 60 * 60 (audit-event.ts:47) and 30 * 24 * 60 * 60 (event-log.ts:13).

3. Settings stored in Mongo but never re-read.

• Settings.verifyCronIntervalHours (admin/models/settings.ts:11) is updatable via PUT /settings.
• runner/session-manager.ts:24-27 reads process.env.VERIFY_CRON_INTERVAL_HOURS once at module load. Mongo updates have no effect until pod restart.
• Same for eventRetentionDays (TTL set at index-creation time, never reapplied via collMod).

Remediation

Build a single sisyphus-api/src/config.ts:

function required(name: string): string {
  const v = process.env[name];
  if (!v) throw new Error(`Required env var ${name} not set`);
  return v;
}

export const config = Object.freeze({
  port: parseInt(process.env["PORT"] ?? "8080", 10),
  mongoUri: required("MONGO_URI"),
  dbName: process.env["DB_NAME"] ?? "sisyphus",
  graphId: required("GRAPH_ID"),  // explicit assert — not "" silently
  chronoGraph: { url: required("CHRONO_GRAPH_URL"), timeoutMs: 30_000 },
  chronoStorage: { url: required("CHRONO_STORAGE_URL"), bucket: required("STORAGE_BUCKET") },
  ornn: { url: required("CHRONO_ORNN_URL"), timeoutMs: 15_000 },
  aevatar: { url: required("AEVATAR_API_URL"), scopeId: required("SCOPE_ID") },
  papers: { maxNodes: 50_000, maxConcurrent: 3, compileTimeoutMs: 300_000 },
  schemas: { lruSize: 500 },
  retention: { eventDays: 30, auditDays: 365 },
  verifyCron: { intervalHours: 6 },  // initial; reload from Mongo on boot
});

Drop the process.env[...] reads in clients; import config.x.

For Mongo-backed settings, on boot read Settings.verifyCronIntervalHours and override config.verifyCron.intervalHours. Add a settings-changed hook (Mongo change stream or polled by version) that calls stopVerifyCron(); startVerifyCron();. For expireAfterSeconds on event_logs, log + reapply via collMod if drifted.

Strip trailing slashes from URLs in the config loader so callers don't have to.