DATA/deerflow-factory
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Initial commit: hardened DeerFlow factory

Browse Source 
Vendored deer-flow upstream (bytedance/deer-flow) plus prompt-injection
hardening:

- New deerflow.security package: content_delimiter, html_cleaner,
  sanitizer (8 layers — invisible chars, control chars, symbols, NFC,
  PUA, tag chars, horizontal whitespace collapse with newline/tab
  preservation, length cap)
- New deerflow.community.searx package: web_search, web_fetch,
  image_search backed by a private SearX instance, every external
  string sanitized and wrapped in <<<EXTERNAL_UNTRUSTED_CONTENT>>>
  delimiters
- All native community web providers (ddg_search, tavily, exa,
  firecrawl, jina_ai, infoquest, image_search) replaced with hard-fail
  stubs that raise NativeWebToolDisabledError at import time, so a
  misconfigured tool.use path fails loud rather than silently falling
  back to unsanitized output
- Native client back-doors (jina_client.py, infoquest_client.py)
  stubbed too
- Native-tool tests quarantined under tests/_disabled_native/
  (collect_ignore_glob via local conftest.py)
- Sanitizer Layer 7 fix: only collapse horizontal whitespace, preserve
  newlines and tabs so list/table structure survives
- Hardened runtime config.yaml references only the searx-backed tools
- Factory overlay (backend/) kept in sync with deer-flow tree as a
  reference / source

See HARDENING.md for the full audit trail and verification steps.
This commit is contained in:
DATA
2026-04-12 14:23:57 +02:00
commit 6de0bf9f5b
889 changed files with 173052 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1306
deer-flow/frontend/public/demo/threads/d3e5adaf-084c-4dd5-9d29-94f1d6bccd98/thread.json Normal file
View File

File diff suppressed because one or more lines are too long

372
deer-flow/frontend/public/demo/threads/d3e5adaf-084c-4dd5-9d29-94f1d6bccd98/user-data/outputs/diana_hu_research.md Normal file
View File

@@ -0,0 +1,372 @@
# Diana Hu: Technical Startup Founder Advice - Comprehensive Research
## Video Overview
**Title:** Tips For Technical Startup Founders | Startup School
**Speaker:** Diana Hu, Y Combinator Group Partner
**Date:** April 21, 2023
**Length:** 28 minutes
**YouTube URL:** https://www.youtube.com/watch?v=rP7bpYsfa6Q
## Speaker Background
### Education
- **BS and MS in Electrical and Computer Engineering** from Carnegie Mellon University
- Focus on **computer vision and machine learning**
- Originally from Chile
### Career Path
1. **Co-founder & CTO of Escher Reality** (YC S17)
- Startup building augmented reality SDK for game developers
- Company acquired by Niantic (makers of Pokémon Go) in February 2018
2. **Director of Engineering at Niantic**
- Headed AR platform after acquisition
- Responsible for scaling AR infrastructure to millions of users
3. **Group Partner at Y Combinator** (Current)
- Has conducted **over 1,700 office hours** across 5 batches
- Advises top YC alumni companies
- Specializes in technical founder guidance
### Key Achievements
- Successfully built and sold AR startup to Niantic
- Scaled systems from prototype to millions of users
- Extensive experience mentoring technical founders
## Escher Reality Acquisition
- **Founded:** 2016
- **Y Combinator Batch:** Summer 2017 (S17)
- **Product:** Augmented Reality backend/SDK for cross-platform mobile AR
- **Acquisition:** February 1, 2018 by Niantic
- **Terms:** Undisclosed, but both co-founders (Ross Finman and Diana Hu) joined Niantic
- **Technology:** Persistent, cross-platform, multi-user AR experiences
- **Impact:** Accelerated Niantic's work on planet-scale AR platform
## Video Content Analysis
### Three Stages of Technical Founder Journey
#### Stage 1: Ideating (0:00-8:30)
**Goal:** Build a prototype as soon as possible (matter of days)
**Key Principles:**
- Build something to show/demo to users
- Doesn't have to work fully
- CEO co-founder should be finding users to show prototype
**Examples:**
1. **Optimizely** (YC W10)
- Built prototype in couple of days
- JavaScript file on S3 for A/B testing
- Manual execution via Chrome console
2. **Escher Reality** (Diana's company)
- Computer vision algorithms on phones
- Demo completed in few weeks
- Visual demo easier than explaining
3. **Remora** (YC W21)
- Carbon capture for semi-trucks
- Used 3D renderings to show promise
- Enough to get users excited despite hard tech
**Common Mistakes:**
- Overbuilding at this stage
- Not talking/listening to users soon enough
- Getting too attached to initial ideas
#### Stage 2: Building MVP (8:30-19:43)
**Goal:** Build to launch quickly (weeks, not months)
**Key Principles:**
1. **Do Things That Don't Scale** (Paul Graham)
- Manual onboarding (editing database directly)
- Founders processing requests manually
- Example: Stripe founders filling bank forms manually
2. **Create 90/10 Solution** (Paul Buchheit)
- Get 90% of value with 10% of effort
- Restrict product to limited dimensions
- Push features to post-launch
3. **Choose Tech for Iteration Speed**
- Balance product needs with personal expertise
- Use third-party frameworks and APIs
- Don't build from scratch
**Examples:**
1. **DoorDash** (originally Palo Alto Delivery)
- Static HTML with PDF menus
- Google Forms for orders
- "Find My Friends" to track deliveries
- Built in one afternoon
- Focused only on Palo Alto initially
2. **WayUp** (YC 2015)
- CTO JJ chose Django/Python over Ruby/Rails
- Prioritized iteration speed over popular choice
- Simple stack: Postgres, Python, Heroku
3. **Justin TV/Twitch**
- Four founders (three technical)
- Each tackled different parts: video streaming, database, web
- Hired "misfits" overlooked by Google
**Tech Stack Philosophy:**
- "If you build a company and it works, tech choices don't matter as much"
- Facebook: PHP → HipHop transpiler
- JavaScript: V8 engine optimization
- Choose what you're dangerous enough with
#### Stage 3: Launch Stage (19:43-26:51)
**Goal:** Iterate towards product-market fit
**Key Principles:**
1. **Quickly Iterate with Hard and Soft Data**
- Set up simple analytics dashboard (Google Analytics, Amplitude, Mixpanel)
- Keep talking to users
- Marry data with user insights
2. **Continuously Launch**
- Example: Segment launched 5 times in one month
- Each launch added features based on user feedback
- Weekly launches to maintain momentum
3. **Balance Building vs Fixing**
- Tech debt is totally fine early on
- "Feel the heat of your tech burning"
- Fix only what prevents product-market fit
**Examples:**
1. **WePay** (YC company)
- Started as B2C payments (Venmo-like)
- Analytics showed features unused
- User interviews revealed GoFundMe needed API
- Pivoted to API product
2. **Pokémon Go Launch**
- Massive scaling issues on day 1
- Load balancer problems caused DDoS-like situation
- Didn't kill the company (made $1B+ revenue)
- "Breaking because of too much demand is a good thing"
3. **Segment**
- December 2012: First launch on Hacker News
- Weekly launches adding features
- Started with Google Analytics, Mixpanel, Intercom support
- Added Node, PHP, WordPress support based on feedback
### Role Evolution Post Product-Market Fit
- **2-5 engineers:** 70% coding time
- **5-10 engineers:** <50% coding time
- **Beyond 10 engineers:** Little to no coding time
- Decision point: Architect role vs People/VP role
## Key Concepts Deep Dive
### 90/10 Solution (Paul Buchheit)
- Find ways to get 90% of the value with 10% of the effort
- Available 90% solution now is better than 100% solution later
- Restrict product dimensions: geography, user type, data type, functionality
### Technical Debt in Startups
- **Early stage:** Embrace technical debt
- **Post product-market fit:** Address scaling issues
- **Philosophy:** "Tech debt is totally fine - feel the heat of your tech burning"
- Only fix what prevents reaching product-market fit
### MVP Principles
1. **Speed over perfection:** Launch in weeks, not months
2. **Manual processes:** Founders do unscalable work
3. **Limited scope:** Constrain to prove core value
4. **Iterative validation:** Launch, learn, iterate
## Companies Mentioned (with Context)
### Optimizely (YC W10)
- A/B testing platform
- Prototype: JavaScript file on S3, manual execution
- Founders: Pete Koomen and Dan Siroker
- Dan previously headed analytics for Obama campaign
### Remora (YC W21)
- Carbon capture device for semi-trucks
- Prototype: 3D renderings to demonstrate concept
- Captures 80%+ of truck emissions
- Can make trucks carbon-negative with biofuels
### Justin TV/Twitch
- Live streaming platform → gaming focus
- Founders: Justin Kan, Emmett Shear, Michael Seibel, Kyle Vogt
- MVP built by 4 founders (3 technical)
- Hired overlooked engineers from Google
### Stripe
- Payment processing API
- Early days: Founders manually processed payments
- Filled bank forms manually for each transaction
- Classic "do things that don't scale" example
### DoorDash
- Originally "Palo Alto Delivery"
- Static HTML with PDF menus
- Google Forms for orders
- "Find My Friends" for delivery tracking
- Focused on suburbs vs metro areas (competitive advantage)
### WayUp (YC 2015)
- Job board for college students
- CTO JJ chose Django/Python over Ruby/Rails
- Prioritized iteration speed over popular choice
- Simple, effective tech stack
### WePay (YC company)
- Started as B2C payments (Venmo competitor)
- Pivoted to API after user discovery
- GoFundMe became key customer
- Example of data + user interviews driving pivot
### Segment
- Analytics infrastructure
- Multiple launches in short timeframe
- Started with limited integrations
- Added features based on user requests
- Acquired by Twilio for $3.2B
### Algolia
- Search API mentioned as YC success
- Part of Diana's network of advised companies
## Actionable Advice for Technical Founders
### Immediate Actions (Week 1)
1. **Build clickable prototype** (Figma, InVision) in 1-3 days
2. **Find 10 potential users** to show prototype
3. **Use existing tools** rather than building from scratch
4. **Embrace ugly code** - it's temporary
### Tech Stack Selection
1. **Choose familiarity over trendiness**
2. **Use third-party services** for non-core functions
3. **Keep infrastructure simple** (Heroku, Firebase, AWS)
4. **Only build what's unique** to your value proposition
### Hiring Strategy
1. **Don't hire too early** (slows you down)
2. **Founders must build** to gain product insights
3. **Look for "misfits"** - overlooked talent
4. **Post product-market fit:** Scale team strategically
### Launch Strategy
1. **Launch multiple times** (weekly iterations)
2. **Combine analytics with user interviews**
3. **Balance feature development with bug fixes**
4. **Accept technical debt** until product-market fit
### Mindset Shifts
1. **From perfectionist to pragmatist**
2. **From specialist to generalist** (do whatever it takes)
3. **From employee to owner** (no task beneath you)
4. **From certainty to comfort with ambiguity**
## Diana's Personal Insights
### From Her Experience
- "Technical founder is committed to the success of your company"
- "Do whatever it takes to get it to work"
- "Your product will evolve - if someone else builds it, you miss key learnings"
- "The only tech choices that matter are tied to customer promises"
### Common Traps to Avoid
1. **"What would Google do?"** - Building like a big company too early
2. **Hiring to move faster** - Actually slows you down initially
3. **Over-fixing vs building** - Focus on product-market fit first
4. **Building features without user insights** - Keep talking to users
## Resources & References
### YC Resources
- Y Combinator Library: "Tips for technical startup founders"
- Paul Graham Essay: "Do Things That Don't Scale"
- Paul Buchheit Concept: "90/10 Solution"
- Startup School: Technical founder track
### Tools Mentioned
- **Prototyping:** Figma, InVision
- **Analytics:** Google Analytics, Amplitude, Mixpanel
- **Infrastructure:** Heroku, Firebase, AWS, GCP
- **Authentication:** Auth0
- **Payments:** Stripe
- **Landing Pages:** Webflow
### Further Reading
1. Paul Graham essays (paulgraham.com)
2. Y Combinator Startup School materials
3. Case studies: Stripe, DoorDash, Segment early days
4. Technical debt management in startups
## Key Takeaways
### For Technical Founders
1. **Speed is your superpower** - Move faster than established companies
2. **Embrace imperfection** - Good enough beats perfect when speed matters
3. **Stay close to users** - Insights come from conversations, not just data
4. **Tech debt is a feature, not a bug** - Early stage startups should accumulate it
### For Startup Strategy
1. **Constrained focus** leads to better unit economics (DoorDash example)
2. **Manual processes** create customer intimacy and learning
3. **Continuous launching** builds momentum and feedback loops
4. **Break things at scale** is a good problem to have
### For Team Building
1. **Founders build first** - Critical for product insights
2. **Hire for adaptability** over pedigree
3. **Evolve role with growth** - Coding time decreases with team size
4. **Culture emerges** from early team composition
---
_Research compiled from YouTube transcript, web searches, and Y Combinator resources. Last updated: January 25, 2026_