Sutskever-Agent/agent.yaml at main · pageman/Sutskever-Agent · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
spec_version: "0.1.0"
name: sutskever-agent
version: 0.1.0
description: >
  A pedagogical AI agent layered on the sutskever-30-implementations repository.
  Helps learners understand, implement, and explore the 30 foundational deep learning
  papers from Ilya Sutskever's reading list — all in pure NumPy.
author: pageman
license: educational
model:
  preferred: claude-sonnet-4-6
  fallback:
    - claude-haiku-4-5-20251001
  constraints:
    temperature: 0.3
    max_tokens: 8192
skills:
  - explain-paper
  - explain-notebook
  - compare-papers
  - review-contribution
  - generate-exercises
  - summarize-repo
  - trace-gradient
tools:
  - github-reader
  - notebook-parser
  - arxiv-fetcher
agents:
  math-explainer:
    description: Breaks down mathematical derivations and notation into accessible steps
    delegation:
      mode: auto
      triggers:
        - math_notation_detected
        - derivation_requested
  code-reviewer:
    description: Reviews NumPy implementations for correctness and pedagogical clarity
    delegation:
      mode: auto
      triggers:
        - code_review_requested
        - pr_opened
runtime:
  max_turns: 40
  timeout: 300
delegation:
  mode: auto
compliance:
  risk_tier: low
  supervision:
    human_in_the_loop: conditional
    kill_switch: true
  recordkeeping:
    audit_logging: true
    retention_period: 1y
    immutable: false
tags:
  - education
  - deep-learning
  - numpy
  - research-papers
  - sutskever
metadata:
  source_repo: https://github.com/pageman/sutskever-30-implementations
  notebook_count: 30
  framework: numpy-only