Review Project 1 provider abstraction

1. 01

   Concepts & system design

   60 min· read · map the mechanismtimer

   R06 — Ollama security

   R11 — Model provenance

   R12 — Safetensors

   Read the assigned sections; map the mechanism, threat model and trade-offs.

2. 02

   Project implementation

   120 min· build · integrate · debugtimer

   Audit P1 provider abstraction, model configuration, artifact provenance, safe loading, and network exposure. Integrate `C8` (Incident diagnosis & recovery recommendations): allow P1 to inspect a restricted read-only slice of structured logs/traces/health signals, classify failures, and propose human-approved recovery steps.

   Build, integrate, debug and improve the increment in the active project.

3. 03

   Security lab

   60 min· attack your own systemtimer

   Verify C8 diagnostic tools are strictly read-only and cannot execute shell commands or mutate state without explicit approval.

   Test the abuse case, failure path, authorization bypass or resource limit.

4. 04

   Verification & evidence

   30 min· test · record evidencetimer

   Pinned P1 model manifest + C8 read-only diagnostic boundary tests.

   Add or run regression tests; record the 9-point evidence record.

5. 05

   Deliberate practice & Socratic review

   30 min· explain aloud · update gapstimer

   Answer the 8 Socratic questions without notes: What? Why? How? What if? Why not? How can I prove it? Where can it break? What have I learned?

   Re-explain the hardest concept aloud, inspect one edge case, update the gap log.

6. total5 h

• R06Ollama Documentation↗ docs.ollama.com

  API endpoints, model management, Modelfiles, and local runtime network binding security.

• R11Hugging Face Documentation↗ huggingface.co

  Model cards, tokenizer/config files, embedding usage, and repository commit-hash revision pinning.

• R12Safetensors Documentation↗ huggingface.co

  Zero-copy safe tensor format overview vs. unsafe pickle deserialization risks.

Read only the named sections needed for today's task. Reading sits inside the 60-minute concepts block; it is not extra time.

• docsOllama · docs
  Ollama Documentation

  Documents the API surface, model management, and Modelfile behaviour, plus the OLLAMA_HOST binding default — the network-exposure check in the P1 audit starts here.

  R06checked 2026-11-11
• videoTech With Tim · 14:01
  Learn Ollama in 15 Minutes - Run LLM Models Locally for FREE

  Demonstrates the local HTTP server and Python client against a running daemon, which is exactly the surface you probe for unauthenticated /api/generate and /api/pull access.

  R06checked 2026-11-11
• videoLeon van Zyl · 14:22
  Run AI Models Locally: Ollama Tutorial (Step-by-Step Guide + WebUI)

  Covers Modelfile authoring and the API endpoints step by step, giving the artifact-provenance trail (which base model, which parameters) that the provider abstraction must record.

  R06checked 2026-11-11
• articleSnyk Labs · technical article
  Vulnerabilities in Deep Learning File Formats

  Walks pickle deserialization to arbitrary code execution and contrasts it with safetensors' raw-tensor-only layout, which is the safe-loading argument the audit has to make per artifact format.

  R12R11checked 2026-11-11
• repoProtect AI (GitHub) · CLI + library
  modelscan

  Scans H5, pickle, and SavedModel artifacts for unsafe operators before load, giving the P1 audit a runnable gate for the model-loading path rather than a policy statement.

  R12checked 2026-11-11

A lab is complete only when it produces evidence. Installing PyRIT, Garak, Presidio or TransformerLens is not completion. Saved on this device only.