## User Feedback Analysis — 2026-04-16 (elizaOS)

### Data coverage note
Feedback available for this date is dominated by GitHub issues/PR discussions and automated code review notes (Greptile/CodeRabbit). Discord daily files referenced in the aggregation are not present, so community sentiment and “how-to” questions in chat are likely underrepresented.

---

## 1) Pain Point Categorization (Top recurring 5–7)

### 1) **Technical Functionality** — Core runtime/memory correctness regressions (high severity)
**Recurring problems (seen in PR review feedback on core changes):**
- **Memory persistence edge-cases create silent correctness loss**
  - IGNORE-path memory persistence bug when `ALLOW_MEMORY_SOURCE_IDS` is configured (responses silently dropped from memory).
  - Embedding failure fallback persists **zero vectors**, creating “stored but unretrievable” semantic memory.
- **Behavioral changes that can feel like regressions**
  - Default provider total timeout changes can push P99 latency upward (noted as potentially breaking for latency-sensitive deployments).

**Why it affects many users:** these are core-path behaviors (message service + runtime memory) that touch nearly every deployment using memory, evaluators, or providers.

---

### 2) **Documentation / Onboarding** — Install & “first project” creation failures (high severity, high friction)
**Recurring problems:**
- `elizaos create <project>` failing on macOS due to **Bun postinstall not run** (Issue #6704). The CLI then cleans up the directory, leaving the user with “nothing to debug.”
- Confusing dependency expectations: `bun` being pulled as a runtime dependency vs being assumed system-installed.

**Impact:** this blocks new users at minute 1. Even a single well-documented blocker here tends to generate repeated support load.

---

### 3) **Integration** — Local dev workflow / submodules & workspace resolution (medium–high severity)
**Recurring problems (from PR #6702 review feedback):**
- Submodule-based plugin dev workflow can break fresh clones if workspace paths are committed in a state that assumes submodules are initialized.
- Lockfile/manifest mismatches (`workspace:*` vs resolved `alpha` dist-tag in `bun.lock`) create non-reproducible installs.

**Impact:** affects contributors, plugin authors, and anyone trying to run the repo from source—i.e., the segment most likely to become maintainers.

---

### 4) **UX/UI (Developer UX)** — Group chat addressee routing & anti-loop behavior is fragile (medium severity)
**Recurring problems:**
- Group-room routing is a major real-world use case (Discord connectors, reply threads).
- A P1 logic bug was identified in addressee resolution for agents with separate `entityId`/`agentId` (PR #6712): aliasing can make name tokens ambiguous, causing addressee detection to return null and routing to fail.

**Impact:** shows up as “agent responds when it shouldn’t” or “agent stays silent when mentioned,” which users experience as unreliability rather than a bug.

---

### 5) **Performance / Reliability** — Async actions and continuation loops (medium severity)
**Recurring problems:**
- Async task actions (PTY-backed coding tasks, agent spawning) can trigger continuation loops that spam filler responses while background work runs (fixed in PR #6709).
- Provider timeouts and response-loop controls are a recurring stabilization theme across PRs.

**Impact:** noisy output degrades trust and raises operational cost in chat-based deployments.

---

### 6) **Community / Ecosystem Quality** — Plugin proposals arrive “feature-rich but unsafe/unfinished” (medium severity)
**Recurring problems:**
- Plugin PRs/proposals include ambitious economic/financial functionality, but often lack:
  - persistence strategy,
  - robust validation,
  - memory bounds/eviction,
  - test coverage,
  - clear threat models (particularly for finance).
- Example: `plugin-mnemopay` PR #6701 flagged for in-memory-only state, NaN corruption risk, ambiguous action matching, and no tests.

**Impact:** review burden increases; users install plugins that may not behave safely, especially in finance-adjacent contexts.

---

### Frequency snapshot (from available GitHub “top issues” on this interval)
- **Economic/marketplace/financial plugins & agent economies:** 3 of 5 top issues (60%) — MAXIA marketplace (#6700), SafeAgent (#6706), AIGEN protocol (#6708).
- **Authorization/identity layer requests:** 1 of 5 (20%) — capability token enforcement (#6707).
- **Onboarding/install blocker:** 1 of 5 (20%) — Bun postinstall create failure (#6704).

---

## 2) Usage Pattern Analysis (Actual vs intended)

### What users are actually building
1) **Agents that move money / trade / coordinate economically**
   - Cross-chain swaps, escrow payments, token safety checks, GPU rental, on-chain rewards.
   - This aligns with the project’s strategic shift toward cryptographic identity + economic layer, but increases the need for formal authorization + auditability.

2) **Agents embedded in group chats (Discord-like connectors)**
   - Real-world usage involves reply threads, multi-party rooms, and “don’t respond to other bots” constraints.
   - The “toon” encapsulation path and routing logic are therefore critical, not edge features.

3) **Long-running, async workflows**
   - Coding tasks, spawned agents, terminal sessions—users expect the agent to “go work” and then return a single clean result.

### Emerging / unexpected use cases
- **AI-to-AI marketplaces as first-class integrations** (MAXIA): agents buying/selling services via escrow, not just calling tools.
- **“Economic memory” and reputation** (MnemoPay): persistent outcomes influencing future decisions, a step beyond conversational memory.

### Feature requests that align with observed usage
- **Capability-based authorization + cryptographic proof** (SINT proposal #6707) matches the reality of high-stakes tool calls (trades, posts, robotics).
- **Pre-trade risk controls** (SafeAgent #6706) matches increased on-chain activity and user safety expectations.

---

## 3) Implementation Opportunities (2–3 concrete solutions per major pain point)

### A) Core runtime/memory correctness (IGNORE persistence, zero-vector embeddings)
**High impact / Medium difficulty**
1) **Add invariant tests for memory write paths**
   - Unit tests covering: normal response, IGNORE response, with/without `DISABLE_MEMORY_CREATION`, with/without `ALLOW_MEMORY_SOURCE_IDS`.
   - Similar approach: LangChain/LLM tool runtimes increasingly rely on regression suites around “message history serialization and replay.”

2) **Change embedding failure policy: “don’t persist semantic index if embedding failed”**
   - Store the raw memory record, but mark embedding as `pending/failed` and exclude from vector search until a successful embedding is computed.
   - Similar approach: many vector DB pipelines (e.g., typical RAG ingestion flows) treat embeddings as a separate, retryable job rather than writing a sentinel vector.

3) **Make timeout default changes explicit + versioned**
   - If defaults change (e.g., provider total timeout), emit a startup warning: “default changed from X→Y; set PROVIDERS_TOTAL_TIMEOUT_MS to keep prior behavior.”
   - Similar approach: Kubernetes deprecation warnings and Next.js config-change warnings reduce surprise regressions.

---

### B) Onboarding failure: `elizaos create` Bun postinstall issue (Issue #6704)
**High impact / Low–Medium difficulty**
1) **Remove `bun` from runtime dependencies; keep only types in dev deps**
   - Implement the proposed fix: move to `@types/bun` and document Bun as an external prerequisite when installing via Bun.

2) **Fail “softly” and keep the generated project directory**
   - If build fails, keep files and print a “Next steps” block with exact commands and logs path.
   - Similar approach: `create-next-app` and many scaffolding tools preserve output and provide remediation steps.

3) **Add an install-time preflight**
   - Detect macOS + Bun + transitive bun package mismatch; warn before project generation and offer `--skip-build`.
   - Similar approach: CLIs like Prisma run environment prechecks and provide targeted diagnostics.

---

### C) Dev workflow fragility: submodules/workspaces/lockfile mismatches (PR #6702 feedback)
**Medium–High impact / Medium difficulty**
1) **Make submodules optional and non-blocking by default**
   - Default install should not require plugin submodules; gated behind a `DEV_PLUGINS=1` or `bun run dev:plugins`.
   - Ensure root `package.json` is always in a clone-safe state.

2) **CI check: “fresh clone install”**
   - Add a CI job that does: `git clone --depth=1`, no submodule init, `bun install`, minimal build.
   - This catches the exact class of “works on my machine with submodules” regressions.

3) **Lockfile consistency enforcement**
   - Pre-commit/CI check ensuring `workspace:*` dependencies resolve as workspaces (or are not present) and that `bun.lock` matches.
   - Similar approach: many monorepos enforce lockfile integrity via CI (pnpm’s `--frozen-lockfile`, etc.).

---

### D) Group addressee routing correctness (PR #6712 alias ambiguity)
**High impact / Medium difficulty**
1) **Fix aliasEntity resolution semantics**
   - Ensure token→entity mapping does not create multi-entity ambiguity when `agentId !== entityId`.
   - Add a deterministic preference rule (e.g., entityId wins; alias stored separately but doesn’t pollute token map).

2) **Add regression tests specifically for entityId/agentId divergence**
   - Current tests don’t cover the failure mode; add fixtures with multiple agents, same nicknames, and distinct IDs.

3) **Expose routing decisions in debug logs**
   - When in group rooms, log: detected addressee tokens, resolved entity, override decision path.
   - Similar approach: Discord bots often provide “why did I respond?” debug toggles to reduce admin confusion.

---

### E) Plugin ecosystem quality (economic/finance plugins especially)
**Medium impact / Medium difficulty**
1) **Introduce a “plugin maturity checklist” for registry inclusion**
   - Required: persistence story (if stateful), bounded memory, input validation, tests, threat model (if financial).
   - Similar approach: Homebrew formula guidelines / VSCode extension marketplace policies.

2) **Provide a reference “stateful plugin” template**
   - Example plugin with: DB adapter usage, migrations, eviction strategy, config parsing with safe defaults.
   - This reduces repeated mistakes (in-memory-only, NaN parsing, null service deref).

3) **Security posture for finance tool calls**
   - Recommend (or optionally require) an authorization gate for high-risk actions (swap/transfer/post).
   - Comparable pattern: wallet dapps commonly enforce confirmation steps; agent frameworks can implement “policy-as-code” gates.

---

## 4) Communication Gaps (Expectations vs reality)

### Gaps observed
- **“ElizaOS is ready for autonomous finance” vs current safety controls**
  - Users are proposing multi-chain trading, escrow, and marketplaces, but core guidance on authorization, audit trails, and safe defaults is not consistently surfaced.
- **Onboarding expectation: scaffolding tools should work out of the box**
  - `elizaos create` failing and deleting output violates common CLI expectations.

### Recurring questions implied by issues/PRs
- “Why does my agent spam filler responses during long tasks?” (continuation loops; addressed by PR #6709)
- “How do I prevent the agent from performing irreversible actions without approval?” (capability token enforcement proposal #6707)
- “Why doesn’t memory persist / why does recall degrade?” (memory controls + embedding correctness concerns in PR review notes)

### Documentation improvements to align expectations
- Add a **“High-risk actions”** doc page:
  - recommended approval tiers for swaps/transfers/posts,
  - audit logging setup,
  - safe plugin patterns.
- Add a **“Bun + CLI installation matrix”**:
  - macOS/Windows/Linux, bun/npm/pnpm behavior, known pitfalls, and remediation.
- Add **“Group chat behavior guide”**:
  - addressee routing metadata (`replyToEntityId`, threads), anti-loop controls, recommended connector settings.

---

## 5) Community Engagement Insights

### Power users (visible in contributions/PRs)
- **odilitime**: driving core runtime observability, routing, and dev harness improvements (multiple major PRs).
- **NubsCarson**: connector-driven bug discovery and practical messaging fixes (toon params, async action loop behavior).

**What they need:**
- Fast feedback loops (clear merge criteria, targeted regression tests).
- Stable internal APIs around messaging/routing and plugin interfaces.

### Newcomer friction signals
- macOS “create project” failure indicates newcomers are hitting environment/toolchain pitfalls early.
- Plugin authors are eager, but need clearer guardrails (testing expectations, persistence patterns, security expectations).

### Converting passive users to contributors
- Label and maintain **“good first issue: docs + onboarding”** tasks (high leverage, lower risk than runtime changes).
- Provide a **plugin submission rubric** and a lightweight certification badge (e.g., “Verified: tests + persistence + bounded memory”).

---

## 6) Feedback Collection Improvements

### Current channel effectiveness
- **GitHub issues/PRs**: high-quality technical feedback, but skewed toward advanced users and contributors.
- **Discord (missing in provided dataset)**: likely contains most onboarding confusion and “how do I…?” questions, but isn’t captured here.

### Improvements for more structured, actionable feedback
1) **Add a post-scaffold survey hook**
   - After `elizaos create`, print a link: “Did setup work? (30 seconds)” capturing OS, package manager, failure mode.

2) **Standardize issue templates**
   - Separate templates for: install/onboarding, runtime correctness, connector behavior, plugin submission.
   - Require minimal reproduction + environment fields for onboarding/install issues.

3) **Underrepresented segments**
   - Non-core contributors running elizaOS as a library (not in-repo dev).
   - Operators deploying in production (latency, logging volume, compliance/audit needs).
   - Windows users (only indirectly referenced via prior plugin-openrouter checkout issues).

---

## Prioritized High-Impact Actions (next 2–4 weeks)

1) **Fix onboarding blocker(s) and stop deleting failed scaffolds**
   - Implement Issue #6704 dependency fix + preserve generated project on failure.
2) **Add regression tests and harden memory correctness paths**
   - Cover IGNORE persistence with allowlists + embedding failure behavior (no zero-vector persistence).
3) **Make repo dev workflow clone-safe**
   - Ensure submodules are optional by default; add CI “fresh clone install” to prevent workspace/lockfile regressions.
4) **Stabilize group chat routing**
   - Fix alias ambiguity bug in addressee routing and add entityId/agentId divergence tests; add debug logging for routing decisions.
5) **Introduce plugin quality gates for stateful/financial plugins**
   - Publish a plugin checklist + provide a reference stateful plugin template (persistence, bounded memory, tests, safe config parsing).