# ElizaOS Strategic Intel — 2026-04-17

## 1) Data Pattern Recognition (Velocity, Trends, Engagement)

### Development velocity (month-to-date signal)
- **Repo:** `elizaos/eliza` (Apr 1 → May 1 interval snapshot)
  - **PRs:** 4 opened / **3 merged** (75% merge rate)
  - **Issues:** **7 opened**, **1 closed** (net +6 backlog growth)
  - **Active contributors:** **14**
  - **Code churn (from sampled PR set):** **+6,785 / −3,732 lines**, **81 files**, **70 commits**
- **Interpretation**
  - Delivery is **high-leverage but spiky**: a small number of PRs account for large surface-area change (core runtime/messaging/prompting).
  - **Backlog is growing** faster than it’s being closed, driven mainly by onboarding + safety/authorization + ecosystem/plugin proposals.

### Change concentration (where the project is moving)
- **Core runtime/messaging** is the center of gravity:
  - Observability + memory controls landed (merged: PR **#6562**).
  - Dev harness + submodule workflow landed (merged: PR **#6702**).
  - Multi-party/group routing + anti-loop prompting is in-flight (PR **#6712**, has a flagged P1 bug).
  - TOON encapsulation/action params reliability work is in-flight (PR **#6709**; large template migration bundled).
- **Ecosystem direction** remains consistent with earlier weekly summary (Apr 5–11):
  - Shift from “messaging protocol” priorities toward **cryptographic identity / authorization** and **agent financial operations** (wallets, safety scoring, marketplace, economic memory).

### Community engagement patterns (qualitative, based on issue/PR content)
- Inbound proposals cluster around:
  1) **Agent commerce** (marketplace, escrow/payments, economic memory, rewards tokens)
  2) **Safety & authorization** (capability tokens, pre-trade checks)
  3) **DX friction** (CLI create failure on macOS/bun; Windows checkout friction in plugin repos)
- Engagement quality is high (detailed issue writeups, concrete APIs, integration plans), but acceptance criteria and plugin pathways appear under-specified (many proposals lack tests, persistence, or compatibility mapping to core identity model).

---

## 2) User Experience Intelligence (Feedback Themes, Impact, Sentiment)

### High-impact UX pain points (ranked)
1) **Onboarding failure: `elizaos create` breaks on macOS (bun dependency/postinstall)**
   - Issue **#6704** (open)
   - Impact: **Hard-block** for first-time users; CLI cleanup leaves “nothing behind,” increasing confusion and drop-off.
   - Likely scope: dependency hygiene in `@elizaos/cli` and `plugin-bootstrap` (runtime vs dev dependency separation).

2) **Multi-party chat quality: addressee routing & anti-loop behavior**
   - PR **#6712** targets a core UX issue: agents replying when not addressed / ping-pong loops in group rooms.
   - Reviewer tooling flags a **P1 logic bug** (aliasEntity ambiguity) that can silently disable the feature for agents where `agentId ≠ entityId`.
   - Impact: reduces trust in “agent social behavior” in Discord/Telegram-style deployments.

3) **Connector/action reliability for TOON encapsulation**
   - PR **#6709** fixes missing action params in TOON schema and reduces filler spam during async task actions.
   - Impact: improves “actions actually execute” rate and reduces noisy intermediate responses (especially in Discord connector scenarios).

### Product sentiment (inferred from content)
- **Positive pull** toward building “real economic agents” (wallets, marketplaces, token incentives).
- **Rising concern** around safety, authorization, and auditability as capabilities expand (capability tokens, evidence ledger proposals).
- **DX patience is thin**: platform-specific friction (macOS bun install semantics; Windows git issues) is repeatedly surfaced and can quickly dampen adoption.

### Usage vs intended design: gaps surfaced
- **Agents are being deployed in group rooms** now (not just single-user chats), forcing deterministic routing needs beyond pure LLM prompting.
- **Financial tooling is being treated as first-class**; users expect:
  - pre-trade safety checks,
  - explicit authorization gates,
  - durable reputation/memory across restarts,
  - auditable logs.

---

## 3) Strategic Prioritization (Impact vs Risk, Dependencies, Resource Focus)

### Immediate “stop-the-bleed” (next 24–72 hours)
1) **Fix CLI macOS create failure (Issue #6704)**
   - **User impact:** Very high (onboarding hard-block)
   - **Tech risk:** Low–medium (dependency graph + release)
   - **Recommendation**
     - Remove `bun` as a runtime dependency where it’s incorrectly pulled transitively; use `@types/bun` as devDependency.
     - Add a **preflight check** in `elizaos create` that detects broken bun postinstall state and prints a single actionable remediation (and *does not delete* the created directory on failure).

2) **Unblock group addressee routing by fixing PR #6712 P1 bug**
   - **User impact:** High for community deployments (Discord/Telegram), reduces spam/looping
   - **Tech risk:** Medium (touches shouldRespond path)
   - **Critical dependency:** correct `NameVariationRegistry.aliasEntity` behavior
   - **Recommendation**
     - Add tests that cover `agentId ≠ entityId` scenarios (the current suite does not catch the flagged failure mode).
     - Ship behind a feature flag (or conservative default) if behavior differences could surprise existing deployments.

### Reliability + quality improvements (next 1–2 weeks)
3) **Land TOON schema/action param + async action loop fix (PR #6709) with scope control**
   - **User impact:** High (actions executing reliably; reduced filler spam)
   - **Tech risk:** Medium due to large template migration bundled into the PR
   - **Recommendation**
     - If possible, **split**: isolate the schema + terminalActions fix from broad prompt/template migration to reduce regression surface.
     - Add a connector-level metric: **% actions invoked with required params present** (before/after).

4) **Close the loop on memory correctness regressions introduced/flagged in core**
   - From merged PR **#6562** review notes:
     - IGNORE persistence logic bug when `ALLOW_MEMORY_SOURCE_IDS` is set
     - Zero-vector embedding fallback causing silent “stored but not retrievable” memories
     - Provider timeout default change that may inflate P99 latency
   - **User impact:** Medium–high (trust in memory; latency)
   - **Tech risk:** Medium (core message pipeline)
   - **Recommendation**
     - Treat as a **patch release** track: small PR(s), targeted tests, explicit changelog callout.

### Strategic bets (2–6 weeks) — evaluate with clear acceptance criteria
5) **Authorization layer alignment: AgentID + capability tokens**
   - Open proposal: `@sint/eliza-plugin` (Issue #6707)
   - **User impact:** High (safe execution of irreversible actions)
   - **Tech risk:** Medium–high (policy engine, UX approval flows, schema compatibility)
   - **Dependency:** formal AgentID data model + policy enforcement points in runtime
   - **Recommendation**
     - Define a minimal “tool authorization interface” in core (hooks/events) before endorsing any single protocol.
     - Pilot with 2–3 tool classes: **on-chain transfer**, **social posting**, **file ops**; require audit log emission consistency.

6) **Economic memory & marketplaces: tighten plugin bar before inclusion**
   - MnemoPay PR **#6701** has critical gaps (no persistence, NaN corruption risk, null deref risk, no tests).
   - MAXIA marketplace proposal (Issue #6700) is broad-scope (hundreds of endpoints/tools).
   - **Recommendation**
     - Publish a **Plugin Acceptance Checklist** for finance-adjacent plugins:
       - persistence strategy (or explicit statelessness),
       - config validation (NaN/invalid env guards),
       - bounded memory growth,
       - test minimums,
       - threat model notes (especially for trading/payment actions),
       - compatibility with AgentID/auth.

---

## Quantitative KPIs to Start Tracking (to optimize dev efficiency + UX)
- **Onboarding**
  - `elizaos create` success rate by OS/package manager (macOS bun, pnpm, etc.)
  - Time-to-first-successful-agent-run (TTFAR)
- **Runtime quality**
  - shouldRespond accuracy in group channels: false-positive reply rate, ping-pong loop rate
  - Provider timeouts: per-provider p95/p99 + failure rate
- **Action reliability**
  - Required-param fulfillment rate (especially TOON/non-streaming connectors)
  - Async action completion: “single clean final response” rate vs filler spam
- **Memory integrity**
  - % embeddings failing + downstream retrieval success rate
  - Memory write suppression reasons (DISABLE_MEMORY_CREATION / allowlist) distribution

---

## Recommended Resource Allocation (next sprint)
- **50% Core reliability & onboarding**
  - Fix macOS create bun dependency issue (#6704)
  - Patch memory correctness issues (IGNORE persistence + embedding fallback)
  - Stabilize shouldRespond/group routing (fix #6712 bug + tests)
- **30% Connector/action execution UX**
  - Land #6709 in a de-risked shape (split if needed)
  - Add action-param + async-action behavioral metrics
- **20% Strategic architecture**
  - Draft core authorization hook spec (to unblock AgentID/capability-token ecosystem coherently)
  - Publish finance-plugin acceptance checklist + reference implementation patterns