Offensive_AI_CON_2025_Framework

This repository contains a lab-safe reference implementation for the talk "The Machine That Hacks Back — agentic API reconnaissance & verification."

It implements an agentic pipeline that turns a blank URL into a reproducible, verifiable test case in an isolated lab:

• Orchestration/Planner with cost-aware simulated annealing and explainable per-step rationales
• Discovery agent with polite crawling, Probabilistic Endpoint Graph (PEG), and multi-modal fingerprinting (headers/size/latency/TLS)
• Contract inference engine with posterior-like schema inference and active-sampling prioritization
• MCP-style tooling adapters (typed contracts) with differential-execution mode, plus adapters for Nuclei and Burp
• Verifier ensemble with counterfactual validation and multi-axis confidence taxonomy
• Safety layer with a Policy DSL, capability tokens, rate limiting, kill switch, immutable audit logs
• Evidence store with cryptographic provenance chains and redaction
• Drift detector, semantic clustering, and adversarial emulator (WAF/5xx/latency)
• CLI for end-to-end runs and a lab Flask app for demos

Quick start (lab only)

python -m lab_app.app
python -m agentic_api.cli discover --base-url http://127.0.0.1:5000 --policy ./configs/policy.dsl
python -m agentic_api.cli infer
python -m agentic_api.cli plan --policy ./configs/policy.dsl --verify-only
python -m agentic_api.cli run --base-url http://127.0.0.1:5000 --policy ./configs/policy.dsl

How it works (start → finish)

• Provide an endpoint and policy
  • Run discover → infer → plan → run (or run to chain them)
  • Policy enforces host/method allowlists, rate limits, and a kill switch
• Discover live surface and build a PEG
  • Polite HEAD/GET crawl collects status, headers, timing, size; computes fingerprints (headers/latency/size/TLS)
  • Writes redacted evidence and artifacts/peg.json
• Infer a coarse contract
  • Estimate parameter types/requiredness; rank uncertain fields (active-sampling priority)
  • Writes artifacts/inferred_schema.json
• Plan safe verification steps
  • Meta-planner optimizes (information gain − cost − risk) and attaches step rationales
  • Writes artifacts/plan.json
• Execute and verify
  • Adapters run steps (HTTP; optional Nuclei/Burp); differential and counterfactual checks
  • Ensemble evaluator produces confidence taxonomy; audit logger records stages
  • Writes artifacts/evidence_card.json
• Optional analysis
  • Semantic clustering: artifacts/semantic_clusters.json
  • Drift compare: artifacts/drift_report.json
  • Mutations + fuzz-diff: mutations.json, snapshot_*.json, fuzzdiff_report.json
• Optional lab demos
  • Nuclei/Burp adapters produce nuclei_result.json, burp_result.json
  • Chain demo (lab PoCs): chain_evidence.json

Advanced commands

# Semantic clustering over PEG
python -m agentic_api.cli semantic --eps 0.5 --min-samples 2

# Drift compare
python -m agentic_api.cli drift --old artifacts/peg_old.json --new artifacts/peg.json --threshold 0.9

# Mutations from inferred schema
python -m agentic_api.cli mutate --limit 25

# Snapshots and fuzz-diff reconciliation
python -m agentic_api.cli snapshot --limit 25 --output snapshot_old.json
python -m agentic_api.cli snapshot --limit 25 --output snapshot_new.json
python -m agentic_api.cli fuzzdiff --old artifacts/snapshot_old.json --new artifacts/snapshot_new.json

# Nuclei (requires nuclei installed)
python -m agentic_api.cli nuclei --url http://127.0.0.1:5000

# Burp (requires Burp REST API + BURP_API_KEY)
python -m agentic_api.cli burp --action scan --url http://127.0.0.1:5000
python -m agentic_api.cli burp --action status --scan-id <id>

# Safe chain demo against lab PoCs
python -m agentic_api.cli chain --base-url http://127.0.0.1:5000

Tests

pytest -q

Safety notes

• Use only in isolated labs or with explicit, written authorization.
• Policy DSL enforces allowlists, rate limits, and approval gates.
• Evidence is redacted and hashed; no raw payloads stored in central logs.

Artifacts

• artifacts/peg.json, artifacts/inferred_schema.json, artifacts/plan.json, artifacts/evidence_card.json, artifacts/semantic_clusters.json, artifacts/drift_report.json, artifacts/snapshot_*.json, artifacts/fuzzdiff_report.json, artifacts/nuclei_result.json, artifacts/burp_result.json, artifacts/chain_evidence.json, artifacts/audit/audit.jsonl