PROVN/provn.yml at main · ashvinctrl/PROVN · 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
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
# provn.yml — Provn Security Policy
# Place in repo root to configure scanning behavior

version: "1.0"

# enforce: block commits with threats
# warn: show warnings but allow commits
# shadow: log only, never block (testing mode)
mode: enforce

# Scan only these file extensions (empty = scan all)
include_extensions: []

# Skip these directories
exclude_dirs:
  - node_modules
  - .git
  - dist
  - build
  - __pycache__
  - .venv
  - target

# Skip these files
exclude_files:
  - "*.lock"
  - "*.min.js"
  - "*.map"

# Risk tier configuration
# T0: Critical — hard block, no bypass allowed
# T1: High — block, bypass with signed reason
# T2: Medium — warn, developer decides
# T3: Low — log only
tiers:
  T0:
    block: true
    bypass_allowed: false
    examples: "Private keys, database passwords, cloud credentials"
  T1:
    block: true
    bypass_allowed: true
    approvers_required: 0
    examples: "API keys, system prompts, model configs"
  T2:
    block: false
    warn: true
    examples: "High-entropy strings, potential IP patterns"
  T3:
    block: false
    warn: false
    log: true
    examples: "Informational findings, low-confidence matches"

# ─────────────────────────────────────────────────────────────────────────────
# Remote allowlist
# Only allow commits when the git remote origin matches one of these patterns.
# Supports * wildcard. Empty list = allow all remotes (default).
# Uncomment and customise to prevent source code from leaving your org.
# ─────────────────────────────────────────────────────────────────────────────
# allowed_remotes:
#   - "git@github.com:mycompany/*"
#   - "https://github.com/mycompany/*"
#   - "git@gitlab.mycompany.com:*"

# ─────────────────────────────────────────────────────────────────────────────
# Layer configuration
layers:
  regex:
    enabled: true
    # Uses built-in Gitleaks-compatible patterns + any custom_patterns below
    # Custom user-defined patterns — checked after built-in patterns.
    # Add regex rules to catch proprietary code, internal imports, copyright violations, etc.
    # Uncomment examples below or add your own.
    # custom_patterns:
    #   # Catch internal package imports leaking into public repos
    #   - name: internal_import
    #     pattern: 'from corp_internal\.'
    #     tier: T1
    #     confidence: 0.90
    #     description: "Internal package reference — do not commit to public repos"
    #
    #   # Catch copyright / confidential markers
    #   - name: confidential_header
    #     pattern: '(?i)CONFIDENTIAL|PROPRIETARY|DO NOT DISTRIBUTE'
    #     tier: T1
    #     confidence: 0.85
    #     description: "File marked confidential — review before committing"
    #
    #   # Catch hardcoded model configs (like the Anthropic/claw-code incident)
    #   - name: model_config_leak
    #     pattern: '(?i)(system_prompt|base_url|model_id)\s*=\s*["\x27][^"x27]{40,}'
    #     tier: T1
    #     confidence: 0.80
    #     description: "Possible model configuration leak"

  entropy:
    enabled: true
    threshold: 4.5
    min_length: 20
  ast:
    enabled: true
    languages: ["python", "typescript", "javascript"]
    # Variables tracked for taint analysis
    sensitive_vars:
      - system_prompt
      - api_key
      - secret
      - password
      - token
      - private_key
      - credentials
  semantic:
    enabled: true
    # Only triggered when Layer 1/2 confidence is ambiguous (0.4-0.8)
    model: "provn-gemma4-e2b-q4km.gguf"
    endpoint: "http://localhost:8080"
    timeout_ms: 2000
    # If model unavailable, fall back to Layer 1/2 result
    fallback: "layer1"

# Auto-redaction
redaction:
  enabled: true
  strategy: "token_replacement"
  # Show preview before applying
  preview: true
  # Prefix for replacement tokens
  token_prefix: "PROVN_REDACTED_"

# Audit logging
audit:
  enabled: true
  path: ".provn/audit.jsonl"
  # HMAC key for chain integrity (auto-generated on first run)
  hmac_key_path: ".provn/hmac.key"
  # Ed25519 key for bypass signing
  signing_key_path: ".provn/signing.key"

# Inline annotations
annotations:
  # Add "# provn:allow" to skip a specific line
  allow_comment: "provn:allow"
  # Add "# provn:skip-file" to skip entire file
  skip_file_comment: "provn:skip-file"