# ElizaOS Intel — 2026-04-14

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

### Development velocity (month-to-date snapshot: 2026-04-01 → 2026-05-01 window)
- PRs: **4 opened / 3 merged**
- Issues: **7 opened / 1 closed** → **net +6** (support/tech-debt backlog growing)
- Contributors: **14 active**
- Code churn: **6,785 additions / 3,732 deletions across 81 files**, **70 commits**
- Change profile: large foundational/runtime changes (observability, memory gating, routing), plus dev harness/workspace work.

### Trend signals
- **Strategic pivot is consistent and compounding**: identity + authorization + economic tooling is pulling community energy (multiple plugin proposals centered on wallets/markets/safety/capability tokens).
- **Runtime core is under active refactor**, increasing throughput but also raising regression risk:
  - Recent merges touched message pipeline, memory persistence controls, provider timeouts, and developer harness bootstrapping.
- **Backlog pressure is developer-experience + safety critical-path**, not feature breadth:
  - Onboarding failures (macOS bun issue) and workspace/submodule workflow complexity are direct adoption blockers.
  - Safety/authorization asks are escalating as agents gain financial and action-taking capabilities.

### Community engagement patterns (observable from issues/PRs)
- Inbound proposals cluster into 3 themes:
  1) **Market access / financial operations** (MAXIA marketplace, AgentWallet, MnemoPay)
  2) **Safety + authorization** (SINT capability token enforcement, SafeAgent pre-trade checks)
  3) **Incentives** (AIGEN contribution economy)
- Engagement is “high-intent builders” (detailed specs, external repos, willingness to maintain) → indicates strong ecosystem leverage if intake/review paths are streamlined.

---

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

### High-impact pain points (ranked by user impact)
1) **CLI/project creation failure on macOS (Apple Silicon)**
   - Issue: `elizaos create` fails with **“Bun’s postinstall script was not run”** when bun is pulled as a transitive runtime dep.
   - UX impact: **hard block** (user ends with no generated project because CLI cleans up).
   - Likely root: packaging/dep strategy mismatch (bun runtime dependency vs system runtime + types).

2) **Cross-platform setup friction**
   - Windows filesystem conflicts were addressed previously (plugin-openrouter), but overall trend indicates **platform-specific install paths remain fragile**.

3) **Message/runtime behavior correctness risks introduced by recent core hardening**
   - Review-identified correctness hazards in merged core work:
     - **Memory IGNORE-path persistence bug when `ALLOW_MEMORY_SOURCE_IDS` is enabled** (silently drops IGNORE memories).
     - **Zero-vector embedding fallback** (persists unretrievable memories → silent functional data loss).
   - UX impact: “agents feel inconsistent / forgetful” and debugging becomes non-obvious.

4) **Group chat routing + anti-loop behavior is improving but has a P1 correctness bug pending**
   - Group addressee routing PR has a flagged issue: **alias ambiguity can break addressee resolution** when `entityId ≠ agentId`.
   - UX impact: agents respond at wrong times (or fail to respond), especially in multi-agent rooms.

### Feature adoption vs intended design (observed mismatch)
- The project intends to move toward **trustable agent economies**, but core runtime currently lacks a **first-class, opinionated authorization gate** between “LLM decided” and “tool executed.”
- Community is already trying to fill this gap via plugins (SINT) and safety layers (SafeAgent), implying:
  - **Adoption pressure is ahead of core guarantees** (agents executing financial actions now).

### Sentiment (inferred from interaction types)
- Positive: strong builder momentum around economic plugins and identity/authorization.
- Negative risk: onboarding failures + regressions in message/memory behavior can quickly convert enthusiasm into churn.

---

## 3) Strategic Prioritization (Impact × Risk, Dependencies, Resource Allocation)

### Initiative scoring (next 1–2 weeks)

| Initiative | User Impact | Technical Risk | Why now | Critical dependencies |
|---|---:|---:|---|
| Fix macOS `elizaos create` bun dependency failure (#6704) | Very High | Low–Med | Direct onboarding blocker; affects first impression | Packaging changes in `@elizaos/cli` + `plugin-bootstrap`; release cut |
| Patch memory correctness regressions (IGNORE persistence + zero-vector embeddings) | High | Med | Silent failures degrade trust; affects “economic memory / identity” roadmap | MessageService + Runtime embedding pipeline; migration/backfill decision |
| Fix addressee routing ambiguity bug in group routing PR (#6712) | High | Med | Group rooms are a primary deployment mode; wrong responses create spam loops | NameVariationRegistry logic + tests for `entityId≠agentId` |
| Stabilize dev harness/workspace workflow (post-merge fallout from agent/ harness PR #6702) | Med–High | Med | Fresh clones/CI failures kill contributor throughput | Workspace/submodule policy; lockfile correctness |
| Define “Authorization & Evidence” core interface (align AgentID + capability tokens) | Very High | High | Needed before scaling financial/tool plugins safely | AgentID spec, policy evaluation points, event/evidence schema |
| Plugin intake governance for financial/safety plugins (wallet/market/safety) | High | Med | Reduces security/regression risk; enables ecosystem scaling | Registry criteria, security checklist, test harness |

### Immediate recommendations (actionable)

#### A) Stop onboarding churn (DX hotfix lane)
1) **Fix bun dependency strategy** for `@elizaos/cli` and `@elizaos/plugin-bootstrap`:
   - Remove `bun` from runtime `dependencies` (use `@types/bun` in dev deps where needed).
   - Adjust `elizaos create` behavior: **do not delete the generated directory on build failure**; leave artifacts + print remediation steps.
   - Add a preflight check: detect `--ignore-scripts`/pnpm/bun edge cases and emit deterministic guidance.

2) **Add a “New project smoke test” CI job**
   - Matrix: macOS (arm64), Ubuntu, Windows
   - Runs: `elizaos create` → install → build → minimal runtime start.

**Resource ask:** 1 maintainer + 1 contributor, 1–2 days to ship + cut patch release.

#### B) Restore trust in memory and runtime correctness (safety/quality lane)
1) **Patch memory correctness**
   - Align IGNORE-path persistence rules with non-IGNORE response path (do not require impossible `agent_response` allowlist entry).
   - Replace “persist zero-vector” embedding fallback:
     - Prefer “do not persist searchable memory until embedding success” or mark as `embeddingStatus=FAILED` and exclude from vector search.
2) **Add regression tests**
   - Tests covering:
     - `ALLOW_MEMORY_SOURCE_IDS` + IGNORE → memory persisted as intended
     - embedding failure → no corrupt vectors stored (or explicitly excluded)

**Resource ask:** 1 core runtime engineer, 1–2 days including tests.

#### C) De-risk group deployments (routing + anti-loop lane)
1) **Fix aliasEntity ambiguity in addressee resolution** (P1 in #6712) before merge.
2) Extend tests to cover:
   - agentId/entityId mismatch
   - multi-agent similar names
   - reply-thread parent author path

**Resource ask:** 0.5–1 day.

#### D) Convert ecosystem pressure into a coherent “agent authorization” platform (strategic lane)
1) **Define a minimal core “Authorization Hook” interface** (core-owned)
   - Standard interception point for tool calls:
     - input: `{agentId, tool, action, params, resourceUri, riskTier}`
     - output: `{allow|deny|requireApproval, evidenceRef}`
2) **Make plugins implement policy providers rather than hard-fork enforcement**
   - Enables SINT-style capability tokens as one implementation, without locking into a single protocol.
3) **Evidence ledger schema alignment**
   - Tie into AgentID direction: behavioral history + signed intent + audit trail.

**Resource ask:** product+tech spec (1–2 days) + staged implementation (1–2 sprints).

---

## Key Watch Items (Next update should confirm)
- Whether macOS create-flow is unblocked and measured (success rate, time-to-first-agent).
- Whether memory correctness fixes land with tests (prevent silent trust erosion).
- Whether group routing merges without alias ambiguity (multi-agent deployments).
- Whether plugin proposals (MAXIA, SafeAgent, SINT, MnemoPay) are funneled through a lightweight security + quality gate to prevent high-risk integrations from bypassing core guarantees.