trace-spec/.github/workflows/runtime-evidence.yml at main · agentrust-io/trace-spec · 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
# The half of the runtime-evidence corpus the ordinary suite cannot check.
#
# `tests/test_runtime_evidence_vectors.py` runs on every push and never imports
# `agent-manifest`: it checks the schema, the record signatures, the evidence
# shapes and the corpus count. Quote verification needs a TDX verifier, which
# TRACE does not ship and the proposal argues it should not start, so that half
# runs here against `agent-manifest`'s verifier at a pinned commit.
#
# Nothing here is mocked or skipped. A missing verifier checkout or a missing
# capture fails this job, because a skipped test would look like coverage and be
# none, which is the defect the proposal itself is about.
name: runtime-evidence

on:
  push:
    branches: ["main", "feat/**", "fix/**"]
  pull_request:
    paths:
      - "examples/runtime-evidence/**"
      - "schema/trace-claim-v0.3-draft.json"
      - "docs/rfcs/runtime-evidence-profile.md"
      - ".github/workflows/runtime-evidence.yml"
      - "requirements/runtime-evidence.txt"

permissions:
  contents: read

env:
  FORCE_JAVASCRIPT_ACTIONS_TO_NODE24: "true"
  # Pinned, not floating. The grades in the proposal are a claim about what a
  # particular verifier says, so the verifier is a version and not a name.
  AGENT_MANIFEST_COMMIT: 934809709a2815695d65cfacb45dc0a164286046

jobs:
  appraise:
    runs-on: ubuntu-latest

    steps:
      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1

      # Fetched into RUNNER_TEMP rather than checked out into the workspace.
      # `tools/check_dashes.py` walks the repository root and would report every em
      # dash in a foreign tree parked there, and it is not the only instrument here
      # that treats everything under the root as ours.
      - name: Fetch the external verifier at the pinned commit
        run: |
          set -euo pipefail
          git init -q "$RUNNER_TEMP/agent-manifest"
          git -C "$RUNNER_TEMP/agent-manifest" remote add origin \
            https://github.com/agentrust-io/agent-manifest
          git -C "$RUNNER_TEMP/agent-manifest" fetch -q --depth 1 origin \
            "$AGENT_MANIFEST_COMMIT"
          git -C "$RUNNER_TEMP/agent-manifest" checkout -q FETCH_HEAD
          test "$(git -C "$RUNNER_TEMP/agent-manifest" rev-parse HEAD)" = "$AGENT_MANIFEST_COMMIT"

      - uses: actions/setup-python@6d4fd22b20465ccc0f7354476588e8df7c9ca0e0 # v7.1.0
        with:
          python-version: "3.11"

      # cbor2 is not a TRACE dependency. Importing `agent_manifest._tdx_verify`
      # goes through the package `__init__`, which needs it; TDX verification
      # itself needs only `cryptography`.
      #
      # Everything comes from hash-pinned locks, as in ci.yml, and the local
      # package goes in with --no-deps. `cryptography` is pinned by
      # requirements/dev.txt, while the verifier declares `>=42,<51`. A lock bump
      # past that bound will fail this job rather than quietly change what the
      # grades mean, which is the direction to fail in; bump the pinned commit
      # when it happens.
      - name: Install package, dev deps, and the verifier's import requirements
        run: |
          pip install --require-hashes -r requirements/dev.txt
          pip install --require-hashes -r requirements/runtime-evidence.txt
          pip install --no-deps -e .

      - name: Appraise the corpus with the real verifier
        env:
          AGENT_MANIFEST_SRC: ${{ runner.temp }}/agent-manifest/python/src
          TDX_CAPTURE: ${{ runner.temp }}/agent-manifest/python/tests/fixtures/hardware/gcp-tdx-2026-07-21
        run: pytest examples/runtime-evidence/test_appraisal.py