Skip to content

github-outreach-report-2026-03-21.md

Latest commit

 

History

History
203 lines (148 loc) · 8.33 KB

github-outreach-report-2026-03-21.md

File metadata and controls

203 lines (148 loc) · 8.33 KB

SupraWall GitHub Outreach — Automated Research Report

Date: 2026-03-21 Status: ⏸ PAUSED — Awaiting user confirmation before opening any issues GitHub account: wiserautomation PyPI package: langchain-suprawall 0.1.0 ✅ verified

Pre-flight Checks

Check Status Notes
langchain-suprawall on PyPI ✅ Pass v0.1.0 released, MIT license
GitHub login ✅ Pass Logged in as wiserautomation
EU AI Act / SupraWall issues already open ✅ Pass 0 existing issues on any of the 5 repos

Repos Skipped

Repo Stars Reason
BCG-X-Official/agentkit 1,946 Archived March 18, 2026 — read-only, cannot open issues

The 5 Target Repos

Repo 1 — wassim249/fastapi-langgraph-agent-production-ready-template

Repo 2 — vstorm-co/full-stack-ai-agent-template

  • Stars: ~854 ⭐
  • Language: Python + TypeScript
  • Last commit: March 20, 2026 (very active — updated yesterday)
  • Description: Production-ready Full-Stack AI Agent Template — FastAPI + Next.js with 5 AI frameworks (PydanticAI, LangChain, LangGraph, CrewAI, DeepAgent)
  • HITL present: No
  • Issue template: Template A
  • Issues URL: https://github.com/vstorm-co/full-stack-ai-agent-template/issues/new
  • Existing EU AI Act / SupraWall issues: 0

Repo 3 — lancedb/vectordb-recipes

  • Stars: ~933 ⭐
  • Language: Jupyter Notebook (Python)
  • Last commit: March 5, 2026 (active)
  • Description: Resources, examples & tutorials for multimodal AI, RAG and agents using vector search and LLMs — includes several LangChain agent examples
  • HITL present: No
  • Issue template: Template A
  • Issues URL: https://github.com/lancedb/vectordb-recipes/issues/new
  • Existing EU AI Act / SupraWall issues: 0

Repo 4 — huangjia2019/ai-agents

  • Stars: ~483 ⭐
  • Language: Jupyter Notebook (Python)
  • Last commit: November 2025 (active)
  • Description: Introductory examples for building LLM-based AI agents with LangChain (companion repo to a published book on AI agents)
  • HITL present: No
  • Issue template: Template A
  • Issues URL: https://github.com/huangjia2019/ai-agents/issues/new
  • Existing EU AI Act / SupraWall issues: 0

Repo 5 — agentailor/fullstack-langgraph-nextjs-agent

  • Stars: ~89 ⭐
  • Language: TypeScript (LangGraph.js)
  • Last commit: February 2026 (active)
  • Description: Production-ready Next.js template for building AI agents with LangGraph.js. Features MCP integration for dynamic tool loading, human-in-the-loop tool approval, persistent conversation memory with PostgreSQL.
  • HITL present: ✅ Yes — custom LangGraph.js interrupt-based tool approval (approve/deny/modify UI)
  • Issue template: Template B ⚠️ (with adaptation — this is TypeScript/LangGraph.js, not Python)
  • Issues URL: https://github.com/agentailor/fullstack-langgraph-nextjs-agent/issues/new
  • Existing EU AI Act / SupraWall issues: 0

⚠️ Note for Repo 5: The langchain_suprawall PyPI package is Python-only. The Template B code snippet references Python. If you open an issue on this TypeScript repo, consider adapting the message to acknowledge the TypeScript ecosystem and mention that a JS/TS SDK is on the roadmap, rather than showing the Python snippet directly.

Issue Content Ready to Post

Repos 1–4: Template A

Title: EU AI Act compliance gap — missing human oversight and audit logging for agent actions

Body:

## Summary

This starter is missing compliance controls required by the EU AI Act for agents deployed to EU users. High-risk obligations apply from August 2025.

## What's Missing

**Article 14 — Human oversight**
No mechanism for a human to review and approve consequential tool calls (e.g., sending emails, modifying data, financial actions) before they execute.

**Article 12 — Audit logging**
Agent executions aren't logged. High-risk AI systems must maintain logs sufficient to reconstruct events after the fact, with a defined retention period.

**Article 9 — Risk management**
No risk classification for the tools this agent can call. A compliant implementation needs a policy layer that evaluates action risk before execution.

## Suggested Fix

Adding `langchain-suprawall` as middleware implements all three requirements:

```python
from langchain_suprawall import SuprawallMiddleware, RiskLevel

executor = AgentExecutor(
    agent=agent,
    tools=tools,
    middleware=[
        SuprawallMiddleware(
            api_key=os.environ["SUPRAWALL_API_KEY"],
            risk_level=RiskLevel.HIGH,
            require_human_oversight=True,   # Article 14
            audit_retention_days=730,       # Article 12: 2-year retention
        ),
    ],
)

This routes high-risk tool calls for human approval (to Slack or a dashboard — not just a terminal prompt), logs every execution to a tamper-evident audit trail, and generates PDF audit reports for regulatory submissions.

Reference

Happy to contribute a PR adding a compliance section to the README if that would be useful.


---

### Repo 5: Template B (adapted for TypeScript)

**Title:** `Human-in-the-loop approval present but doesn't satisfy EU AI Act Article 14 in production`

**Body:**
```markdown
## Summary

This repo has a great human-in-the-loop tool approval flow — that's a strong start. However, the current implementation has a gap relative to EU AI Act Article 14 for production deployments.

## The Gap

The current HITL implementation routes approval requests to a UI dialog. Article 14 requires approvals to be **logged** with approver identity and timestamp — a UI dialog that doesn't write to a tamper-evident audit trail doesn't satisfy this. Additionally, Article 12 requires a defined log retention period.

## What's Needed for Full Compliance

- Approver identity + timestamp persisted for every approval decision (not just execution logs)
- Tamper-evident audit trail (append-only, cryptographically verifiable)
- Defined retention period (≥ 2 years for high-risk systems)
- Ability to generate audit reports for regulatory submissions

## Suggested Fix (Python equivalent — JS/TS SDK on the SupraWall roadmap)

```python
# After: Slack/dashboard approval with full audit logging (Articles 12 + 14)
from langchain_suprawall import SuprawallMiddleware
hitl = SuprawallMiddleware(
    api_key=os.environ["SUPRAWALL_API_KEY"],
    require_human_oversight=True,
    notification_channel="slack",
    audit_retention_days=730,
)

Reference


---

## Next Steps

To proceed, Alejandro needs to confirm each issue one by one in the chat:

1. ✅ "Open issue on **wassim249/fastapi-langgraph-agent-production-ready-template**?" → 2,100-star Python production template
2. ✅ "Open issue on **vstorm-co/full-stack-ai-agent-template**?" → 854-star very-active full-stack template
3. ✅ "Open issue on **lancedb/vectordb-recipes**?" → 933-star LangChain/RAG examples repo
4. ✅ "Open issue on **huangjia2019/ai-agents**?" → 483-star introductory AI agents examples (book companion)
5. ✅ "Open issue on **agentailor/fullstack-langgraph-nextjs-agent**?" → 89-star LangGraph.js TypeScript template (HITL present, Template B adapted)

Issues will be spaced ≥30 seconds apart to avoid GitHub rate limiting.

---

*Report generated automatically by the suprawall-github-outreach scheduled task.*