Repository navigation
Description
Classification
- Change type: enhancement / gate integration
- Priority: P1
- Expected value: high after replayable run semantics land — lets agent runs collect artifact provenance evidence without putting supply-chain domain logic into 007
- Implementation risk: medium — false-green gates are worse than no gate
- Repository owner: 007 execution/orchestration layer
- Depends on: Run semantics: replayable event reducer and fail-closed required-gate verdicts #43; consumes
PhysShell/OwnAudit#44andPhysShell/Own.NET#271 - Production status: advisory only until Run semantics: replayable event reducer and fail-closed required-gate verdicts #43 makes required gate verdicts fail-closed and replayable
Problem
007 should not become a supply-chain scanner. It already has one job: isolate a target worktree, run an agent, execute gates, harvest evidence and preserve a run record.
Supply/provenance checks belong in OwnAudit as profile-driven evidence adapters. 007 should only invoke an OwnAudit profile and record the resulting artifacts.
Also, 007 currently has a known semantics gap tracked in #43: required gates that do not actually execute must not accidentally reduce to PASS. Until that is fixed, any supply-evidence gate must remain advisory/non-blocking.
Goal
Add a narrow integration path for an OwnAudit supply evidence profile:
[[gate]]
name = "supply-evidence-ci-light"
cmd = "python audit/supply/run.py --profile ci-light --target artifacts/bin --out artifacts/supply"
required = false007 must treat this as just another gate command with artifact capture. It must not know SL001, SC005, SourceLink, NuGet signatures, Scorecard or SBOM semantics.
Required behavior
- The gate is advisory by default.
- The command path/profile is configured in the target repo, not hardcoded in 007.
- 007 records produced artifacts by digest, for example:
artifacts/supply/evidence.jsonl
artifacts/supply/supply.sarif
artifacts/supply/coverage.json
- If Run semantics: replayable event reducer and fail-closed required-gate verdicts #43 event streams are available, the gate result and artifact digests become replay-verifiable run events.
- Missing supply artifacts are visible as a gate/evidence limitation, not silently ignored.
- Required mode is allowed only after fail-closed gate semantics are implemented and tested.
Relationship to OwnAudit
OwnAudit owns supply evidence logic:
nuget-resolution-policy
sourcelink-debuggability
nuget-integrity
known-vuln
sbom
repo-posture
attestations
rebuild-lab
007 owns only:
run command
capture stdout/stderr
capture exit code
capture declared artifacts/digests
reduce/replay gate state
Suggested target gate file
For a target repository:
[[gate]]
name = "supply-evidence-ci-light"
cmd = "python audit/supply/run.py --profile ci-light --target artifacts/bin --out artifacts/supply"
required = false
[gate.artifacts]
paths = [
"artifacts/supply/evidence.jsonl",
"artifacts/supply/supply.sarif",
"artifacts/supply/coverage.json"
]The exact manifest shape may change with #43/#42. The semantic requirement is stable: 007 records the evidence; it does not interpret it.
Non-goals
- No SL/SC rule knowledge in 007.
- No supply-chain policy engine in 007.
- No required/blocking supply gate before Run semantics: replayable event reducer and fail-closed required-gate verdicts #43.
- No sandbox implementation here.
- No rebuild runner here.
- No retry/repair loop here.
Required controls
- Advisory supply gate passes/fails without changing overall required verdict.
- Required supply gate is rejected or blocked until required-gate semantics are fail-closed.
- Gate artifact digest changes after run sealing are caught by replay once Run semantics: replayable event reducer and fail-closed required-gate verdicts #43 lands.
- Missing artifact path is represented in the run evidence.
- Legacy runs without supply artifacts remain readable with explicit limitations.
- A target repo can remove the supply gate without breaking 007.
Acceptance
- Documentation shows how to add a non-required OwnAudit supply profile gate.
- 007 records declared supply artifacts and digests when present.
- 007 does not parse or classify SL/SC findings.
- Required mode is blocked/deferred until Run semantics: replayable event reducer and fail-closed required-gate verdicts #43 closes.
- Replay/event integration is described or implemented through the Run semantics: replayable event reducer and fail-closed required-gate verdicts #43 event contract.
Activity