DEVELOPMENT ROADMAP

Active Items

Develop a multi-agent system for monitoring and managing tokenized funds on-chain.

Build a narrow-scope DeFi yield operations agent modeled on Giza's ARMA (9.75% APY on Aave/Morpho). DWF Ventures data confirms AI agents lose to humans 5:1 at open-ended trading but WIN at narrow operational tasks: yield routing, rebalancing, protocol health monitoring. Target: deploy a Grid agent that continuously monitors DeFi yield rates across Aave, Morpho, and Compound, auto-rebalances between pools when yield differential exceeds a threshold, and tracks position health. Revenue model: performance fee (% of yield above benchmark) + fixed monthly fee per managed position. Start with paper-mode yield simulation before live capital.

Build and launch the Prediction Market Daily Intelligence subscription product. Deliver automated daily reports covering: top 10 liquid markets across Kalshi + Polymarket, probability shifts vs. prior day, cross-platform price dislocations, event resolution signals, and negative-risk-sum arbitrage candidates. Delivery: Telegram channel post + email digest. Pricing: /month OR /report a la carte. The yori-pm-intelligence.mjs script exists but the subscription delivery pipeline (scheduled synthesis + formatted output + subscriber list management) has not been built. Cost: ~-2/report at current research API rates = 85%+ gross margin. Break-even: 7 subscribers.

Audit /api/v1/* for naming inconsistencies (camelCase vs snake_case, plural vs singular). Adopt a single convention. v1.x stays — breaking changes go in v2.

Productize cmc-defi-watchlist as a paid weekly audit report. Targets DAO treasuries and institutional DeFi users. $499/mo.

Productize the existing OpenClaw multi-provider routing logic as a hosted service. Customer brings their own API keys; Grid handles cost-aware routing.

Niche newsletter: tactical PM trading. NEON DAVE writes. Paid tier $39/mo for picks + EV models.

Add 'import { installCronGuard } from "./lib/cron-guard.mjs"; installCronGuard("<task-id>", { timeoutMs: 90_000 });' (or .cjs require if applicable) at the top of scripts/a2a.mjs, just after the shebang. Prevents silent multi-hour wedges per project_cron_guard_infrastructure.md.

Add 'import { installCronGuard } from "./lib/cron-guard.mjs"; installCronGuard("<task-id>", { timeoutMs: 90_000 });' (or .cjs require if applicable) at the top of scripts/activity-timeline.mjs, just after the shebang. Prevents silent multi-hour wedges per project_cron_guard_infrastructure.md.

Add 'import { installCronGuard } from "./lib/cron-guard.mjs"; installCronGuard("<task-id>", { timeoutMs: 90_000 });' (or .cjs require if applicable) at the top of scripts/add-claude-inbox-to-prompts.mjs, just after the shebang. Prevents silent multi-hour wedges per project_cron_guard_infrastructure.md.

Add 'import { installCronGuard } from "./lib/cron-guard.mjs"; installCronGuard("<task-id>", { timeoutMs: 90_000 });' (or .cjs require if applicable) at the top of scripts/add-dumont-agentverse-register.mjs, just after the shebang. Prevents silent multi-hour wedges per project_cron_guard_infrastructure.md.

Add 'import { installCronGuard } from "./lib/cron-guard.mjs"; installCronGuard("<task-id>", { timeoutMs: 90_000 });' (or .cjs require if applicable) at the top of scripts/add-heartbeat-fanout.mjs, just after the shebang. Prevents silent multi-hour wedges per project_cron_guard_infrastructure.md.

Add 'import { installCronGuard } from "./lib/cron-guard.mjs"; installCronGuard("<task-id>", { timeoutMs: 90_000 });' (or .cjs require if applicable) at the top of scripts/add-research-integration-job.mjs, just after the shebang. Prevents silent multi-hour wedges per project_cron_guard_infrastructure.md.

Add 'import { installCronGuard } from "./lib/cron-guard.mjs"; installCronGuard("<task-id>", { timeoutMs: 90_000 });' (or .cjs require if applicable) at the top of scripts/add-roadmap-entry.mjs, just after the shebang. Prevents silent multi-hour wedges per project_cron_guard_infrastructure.md.

Add 'import { installCronGuard } from "./lib/cron-guard.mjs"; installCronGuard("<task-id>", { timeoutMs: 90_000 });' (or .cjs require if applicable) at the top of scripts/add-tools-policy.mjs, just after the shebang. Prevents silent multi-hour wedges per project_cron_guard_infrastructure.md.

Leaked gho_ token in commit 2bd2d1c, redacted in working tree at 960a9bf but still present in history. Repo logikclaw/the-grid is private (verified via API), so blast radius is contained, but the practice is still wrong. Two-step: (1) MC revokes the leaked token via GitHub UI and issues a new fine-grained PAT (one-time human action, alerted via Telegram msg 4230), (2) once new token is in place, run git filter-repo or BFG to remove the token string from history + force-push.

Improve content scheduling algorithms and integrate analytics data directly into the publishing pipeline.

Develop API connections and native applications for emerging social media platforms.

Create interactive features such as polls, quizzes, live Q&A sessions to enhance user engagement.

Today the only way to get a gk_* key is for an MC-side script to mint one (onboard-test-customer.mjs). Need a public flow: gridfleet.ai signup → email verification (via gridfleet-mail Worker — we own *@gridfleet.ai inbound now) → API key issuance + welcome email + Stripe customer creation. Without this no real customer can self-serve.

Static pricing page at gridfleet.ai/pricing. Tiers: Free (100 calls/mo), Starter ($29, 5k), Pro ($99, 50k + captcha), Enterprise ($499, 500k + SLA). Cost-per-call breakdown for the ala-carte tier. Comparison table vs. Browserbase/Browser-Use Cloud. Links to /signup once that exists.

Self-hosted API docs covering POST /v1/services/browse + GET /v1/services/pong/* endpoints. Format: a single static HTML page with curl examples + auth notes + tier rate limits + error codes. Source from the inline GET handler on browse-route.ts (already returns a JSON spec) — render that as a human page. Future-proof: when more endpoints land, regenerate from spec.

When mail to sales@/support@ lands as a GitHub Issue (already wired via gridfleet-mail Worker), DUMONT or YORI is supposed to triage. There's no playbook today — what fields to extract, what reply template to use, what qualifies as a high-intent lead, when to escalate to MC, what SLA to honor. Write workspace/knowledge/shared/playbooks/sales-support-runbook.md with concrete reply templates + decision flowchart.

Page at gridfleet.ai/account where a customer signs in (with their gk_* key as auth) and sees: calls_this_month, usd_this_month, quota_calls, quota_usd, last 50 billing events, current tier, upgrade button. Reuses the existing customer-auth lib + getMonthlyUsage(). No new auth flow needed — just a UI on top of what exists.

Cron polling Aave + Morpho + Compound public APIs every 15 min for current supply APY across stablecoin pools. Append to workspace/state/defi-yield-rates.jsonl. No customer exposure yet.

Reads rate-monitor output, computes which pool offers best APY for a hypothetical $X position, recommends rebalances when differential exceeds threshold. Paper mode only — output goes to workspace/state/defi-yield-recommendations.jsonl.

For each (hypothetical or real) deployed position, compute current LTV vs. liquidation threshold, time-to-liquidation under stress scenarios. Alert when distance < 20%.

What would have happened if we'd executed the rebalance recommendations? Tracks hypothetical P&L vs. buy-and-hold benchmark. Requires N weeks of data before performance fee model is justified.

Add a daily 7am MT cron to run the existing yori-pm-intelligence.mjs script. Output goes to workspace/state/pm-signals-daily/<date>.json.

Take the daily output JSON, format as: (a) Telegram-channel-ready HTML, (b) email-ready HTML+text, (c) markdown for /reports/ public page. Each with the same content but the right escaping.

Create @gridfleet_pm_signals (or similar) Telegram channel + bot poster. Free tier: daily summary. Subscribers also get the full report via DM. Bot tokens already exist.

Send daily report by email to paid subscribers. BLOCKED — no outbound SMTP working today. Unblocks once we land a Brevo/MailerSend/Postmark signup (see infra-hn-autonomous-pipeline notes).

Per knowledge/active/PATH_TO_CODEBASE_9.md: lift coverage on critical paths (cron engine, telegram channel, weather signals, ISO emergence) to 30-40%. Run TS strict, eliminate explicit `any` from src/. Wire to a CI gate.

Tighten internal module boundaries — convert ad-hoc fs.readFileSync calls scattered through routes into a shared lib/state-store with typed accessors. Add integration tests across the boundaries.

Naming inconsistencies (already partially deferred via codebase-quality-4-naming), dead-code sweep, dependency audit, unused-import lint.

Stretch goal. Coverage >= 60%, real observability stack, chaos drills, doc generation from code. Only triggered after 4A-4C land.

GitHub Actions: lint + TypeScript strict + ~30-40% coverage of critical paths. DUMONT is the sole Git authority per constitution. Prevents schema-crash repeats. Closes Phase 2 item 5 of PATH_TO_90_AUDIT_ROADMAP.md.

BECK is wired to Hermes (per memory). Verify the ISO auto-approver path for LOG-SWEEP, CRON-DOCTOR, PIPE-REFRESH classes is actually live and reversible-only. May already be done — just needs a status check + closure note.

Per memory project_thegrid_en_dispatch_broken.md: binding agents to ollama/thegrid-en:latest silently wedges gateway. Requires source-level openclaw fix, not config tuning. BLOCKED on upstream OpenClaw release.

Today most roadmap entries have no acceptance criteria. Add a process check: when add-roadmap-entry.mjs runs, require an AC field with at least one bullet. Backfill existing entries during the next triage pass.

Per memory project_a2a_shipped.md: 5/7 acceptance criteria met. Identify the remaining 2 and close them. Likely candidates: cross-agent message acks + A2A-over-network (vs file-bus only).

Research-only. What's Skyfire's actual API for agent-pays-agent? Pricing? Reservation model? Fee take? Latency? Does it work for our $0.005-0.10 per-call price points or is the floor higher?

Research-only. Does AGNTCY's directory accept our existing MCP servers? Is listing free? Does it have its own payment rails or just discovery?

Define how a paying agent authenticates + pays for an MCP call. Either via Skyfire reservation token, x402-style HTTP 402 + payment header, or a custom Bearer extension. BLOCKED on outcome of the two evaluations above.

The parent entry says only 'Develop a multi-agent system for monitoring and managing tokenized funds on-chain.' That's not actionable. Spend a session: which fund(s)? On which chain(s)? What does 'monitor' mean (NAV drift, redemption queue depth, custodian-key rotation, oracle attestation)? What does 'manage' mean (we are NOT a fund manager)? Output: tighter spec or recommendation to defer.