USE CASES (3 found):

1. [omarshahine] uses OpenClaw for iMessage channel inbound catchup Source: github.com/openclaw/openclaw/pull/79387 Key insight: Building production-grade message recovery for chat.db when gateway restarts/crashes — with cursor management, replay loops, and per-account isolation. Shows OpenClaw is being used for serious Apple ecosystem automation.

2. [easyFloyd] built FadNote as an OpenClaw Skill Source: news.ycombinator.com/item?id=47253272 Key insight: Zero-knowledge secret sharing (AES-256-GCM, URL fragment keys) integrated directly into OpenClaw workflows — proving the skill system works for security tools, not just chat.

3. [Hermes_dev / HalluciGuard] built hallucination detection middleware with native OpenClaw interceptor Source: news.ycombinator.com/item?id=47094986 Key insight: Real-time multi-signal verification (self-consistency, Tavily web search, RAG cross-reference) with a cost-optimization cache. They chose AGPLv3 because they believe the "truth layer" should be community-owned.

CONFIG INSIGHTS (3 found):

1. [soul.md project] defines SOUL.md as identity continuity, not capabilities Source: soul.md / github.com/aaronjmars/soul.md Suggestion for Karl: Include an explicit line in your SOUL.md acknowledging session boundaries: "I don't remember previous sessions unless I read my memory files." This grounds the agent and improves self-awareness.

2. [nami_creator] sells pre-written config packs (SOUL.md + HEARTBEAT.md + AGENTS.md) Source: news.ycombinator.com/item?id=47496354 Suggestion for Karl: Consider versioning your SOUL.md and MEMORY.md with git tags. If your configs grow complex enough that people sell them, yours should be versioned too.

3. [hclsys] found that per-agent string models silently clobber default fallbacks Source: github.com/openclaw/openclaw/pull/79389 Suggestion for Karl: Run openclaw doctor after any model configuration change — the doctor now warns when a per-agent model string overrides your fallback chain silently.

TECHNICAL IMPROVEMENTS (3 found):

1. [HalluciGuard] built a verification pipeline with Tavily web search integration Source: github.com/Hermes-Lekkas/HalluciGuard What it does: Intercepts LLM responses, extracts factual claims, verifies via Tavily + RAG + self-consistency scoring Advantage over default: Provides a "trust score" per claim instead of blind faith in model output Suggestion for Karl: If you do research-heavy tasks, a Tavily integration would give you real-time web snippets with source attribution — better than raw search results.

2. [hclsys] added config.patch support for tools.web.search.maxResults Source: github.com/openclaw/openclaw/pull/79391 What it does: Lets you tune how many search results OpenClaw fetches per web search call Suggestion for Karl: If research tasks feel too shallow or too noisy, patch tools.web.search.maxResults in your config instead of rewriting prompts.

3. [quintusadvisorygroup] reported and characterized a gateway CPU spin bug Source: github.com/openclaw/openclaw/issues/79422 Key feature: Identified CopyDataPropertiesWithExcludedProperties in a hot async microtask loop (object spread) causing 100%+ CPU and 18–19 GB log bloat Suggestion for Karl: If your gateway ever spikes, restart it immediately (openclaw gateway restart) — but also check ~/.openclaw/logs/gateway.err.log size. This bug recurs without a code fix.

No new SOUL.md/MEMORY.md best-practice blog posts today. The HN signal on config tips is still thin — mostly repo links and productized templates rather than deep writeups.

Report complete. No files modified.