Skip to content

Release v2.0: agents, MCP, and the Agentic Kill Chain - #11

Merged
iamfarooqh merged 15 commits into
mainfrom
v2.0-rc-alpha
Oct 10, 2026
Merged

iamfarooqh merged 15 commits into
mainfrom
v2.0-rc-alpha

Conversation

@iamfarooqh

Copy link
Copy Markdown
Member

Summary

AIGoat v2.0 grows the platform from an LLM chatbot target to the surfaces that appear once a model can act. Labs grow from 17 to 46 across three OWASP frameworks (LLM Top 10 2026, Agentic Top 10 2026, MCP Top 10).

What's in this PR

  • Tool-calling agent, MCP host and client with real MCP servers, per-lab agent memory
  • Agentic Kill Chain capstone (killchain-1): review/PDF memory poisoning, dual memory, approvals
  • New Guardrailed mode for the kill chain: ingest quarantine, egress allowlist, card masking, staff-coupon refusal, answer masking
  • Shop prices in the kill chain now match the storefront
  • Unified hub UI, Learn pages, Threat Modeling workbench, Google Colab notebook
  • Alembic migrations (0001 to 0009); start.sh stamps pre-Alembic databases at 0001, then migrates
  • start.sh now checks Node 18+, waits for the API and the app to answer, and checks the tool-calling model

Upgrade notes

  • Lab IDs follow the 2026 numbering (for example llm10-1 is now llm06-1). Migration 0005 renames stored data.
  • Pull a tool-calling model for the agent, MCP and kill chain labs: ollama pull qwen3.5:9b
  • Existing kill chain users press Hard reset once to re-copy shop data.
  • Unchanged: seed users and passwords, the 9 CTF challenges, the flag format, API paths (changes are additive).

Verification

Backend lint and 1137 tests, frontend tests and build, fresh start with start.sh --fresh.
Full notes: RELEASE_NOTES_v2.0.md

iamfarooqh and others added 15 commits September 20, 2026 19:07
…o v2-labs

Co-authored-by: Cursor <cursoragent@cursor.com>

# Conflicts:
#	app/agent/loop.py
#	app/api/labs.py
#	app/challenges/evaluators/mcp_host.py
#	app/core/lab_loader.py
#	app/mcp/host.py
#	config/defense_profiles.yml
#	config/frameworks/owasp-agentic-2026.yml
#	config/labs/agent.yml
#	config/labs/mcp.yml
#	frontend/src/components/admin/AdminAssistantPage.jsx
#	frontend/src/utils/labTeaching.js
#	tests/test_lab_manifest.py
#	tests/test_labs_api.py
#	tests/test_mcp_api.py
- labTeaching: correct stale review-based asi01-1 copy to the actual
  ticket flow (SHIP_CONTROL, MCP_FLOWS diagram, PLANT, ASK_PROMPT)
- McpHubPage/AttacksPage: MCP05 card says 'sandboxed executor lab'
  instead of 'refused executor' since asi05-1 runs in Docker
- LabWorkspace: back button for ASI labs on the mcp.host surface
  returns to the Agent hub instead of the MCP hub

Co-authored-by: Cursor <cursoragent@cursor.com>
Bring MCP, Agent, RAG, OWASP, Attack Labs, Challenges, and Threat Modeling onto one hub layout, replace killchain-1 with the compromised ecommerce agent lab, and keep Docker reachable publicly
Deterministic rails still block a wrong approval: ingest quarantine, an egress allowlist, card masking, and a staff-coupon refusal. Stamp pre-Alembic databases at revision 0001 so later migrations run, and refresh the README, license paths, and guardrails guide for the v2.0 release.
@iamfarooqh
iamfarooqh merged commit e1c538b into main Oct 10, 2026
4 checks passed
nalinikanth pushed a commit that referenced this pull request Oct 18, 2026
Release v2.0: agents, MCP, and the Agentic Kill Chain
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants