quorum/quorum-relationships.example.yaml at main · SharedIntellect/quorum · 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
# quorum-relationships.yaml — Declare relationships between artifacts for cross-validation
#
# Usage:
#   quorum run --target <file> --relationships quorum-relationships.yaml
#
# See docs/CROSS_ARTIFACT_DESIGN.md for full schema documentation.
#
# Supported relationship types:
#   implements      — spec/impl coverage verification
#   documents       — source/docs accuracy verification
#   delegates       — from/to boundary verification
#   schema_contract — producer/consumer compatibility verification
#   threat_context  — security context for authorization review (SEC-04)

relationships:
  # Example 1: Verify the pipeline implements its design spec
  - type: implements
    spec: docs/CROSS_ARTIFACT_DESIGN.md
    impl: quorum/pipeline.py
    scope: "Phase 2 cross-artifact coordination"

  # Example 2: Verify the CLI docs accurately describe the CLI source
  - type: documents
    source: quorum/cli.py
    docs: docs/CONFIG_REFERENCE.md

  # Example 3: Verify code_hygiene delegates security patterns to the security critic
  - type: delegates
    from: quorum/critics/code_hygiene.py
    to: quorum/critics/security.py
    scope: "Security-adjacent patterns (eval/exec, credentials, prompt injection)"

  # Example 4: Verify PreScreenResult model contract between producer and consumer
  - type: schema_contract
    producer: quorum/prescreen.py
    consumer: quorum/critics/security.py
    scope: "PreScreenResult model — to_evidence_block() output format"

  # Example 5: Provide threat model context for authorization review (SEC-04)
  # Without this, the Security Critic evaluates authorization logic in a vacuum —
  # it can spot missing auth checks, but can't assess whether the right roles
  # have access to the right resources.
  - type: threat_context
    target: quorum/pipeline.py
    context:
      roles:
        - name: admin
          description: "Full system access, can configure critics and rubrics"
        - name: operator
          description: "Can trigger validation runs, read results"
        - name: viewer
          description: "Read-only access to verdicts and reports"
      trust_boundaries:
        - name: "CLI → Pipeline"
          description: "User input enters via CLI args and config files"
        - name: "Pipeline → LLM Provider"
          description: "Prompts and artifact content sent to external API"
        - name: "Pipeline → Filesystem"
          description: "Run artifacts written to local disk"
      sensitive_operations:
        - "Rubric loading (arbitrary file read)"
        - "Pre-screen shell execution (Ruff/PSSA invocation)"
        - "LLM prompt construction (prompt injection surface)"
        - "Verdict writing (integrity of assessment output)"