API Task Results

1. Repository Clone — ✅ Success

Endpoint: POST $PROVISIONER_URL/repo/provision

Request:

{
  "repo": "fullsend-ai/fullsend",
  "ref": "main",
  "dest": "/sandbox/fullsend",
  "sandbox": "<sandbox-name>"
}

Result:

• Status: completed
• Operation ID: 27d8cf99-f519-441e-8c1b-3993bca7fe65
• Destination: /sandbox/fullsend/repo/
• Security Findings (all medium, non-blocking):
  • internal/scaffold/fullsend-repo/skills/code-review/SKILL.md — Prompt injection pattern: <SYSTEM>
  • docs/problems/security-threat-model.md — Prompt injection pattern: ignore previous instructions (×2)
  • .git/hooks/pre-commit — Executable git hook

The repo was cloned, scanned, and uploaded successfully despite the medium-severity findings.

---

2. Container Image Build — ✅ Success

Endpoint: POST $BUILDER_URL/build

Request:

{
  "tag": "fullsend-sandbox:test",
  "dockerfile": "images/sandbox/Containerfile",
  "context_dir": "/sandbox/fullsend/repo",
  "dest": "/sandbox/fullsend-sandbox.tar",
  "sandbox": "<sandbox-name>"
}

Result:

• Status: success
• Operation ID: d62a038c-acb1-40c3-9def-549a08b88424
• Tag: fullsend-sandbox:test
• Image tarball: /sandbox/fullsend-sandbox.tar (4.2 GB)
• Build steps: 36/36 completed (all cached layers)
• The Containerfile installs Claude CLI, ProtectAI DeBERTa-v3 model, ONNX Runtime, Gitleaks, pre-commit, Tirith, and ACLI on top of the OpenShell base sandbox image.

---

3. Access Control / 403 Status

No 403 errors were encountered. All endpoints used (/repo/provision, /build) responded successfully.

Endpoints Tested

Server	Endpoint	Method	Status
Provisioner	/openapi.json	GET	✅ 200
Provisioner	/repo/provision	POST	✅ 200
Builder	/openapi.json	GET	✅ 200
Builder	/build	POST	✅ 200

Endpoints Not Tested (available per OpenAPI spec)

Server	Endpoint	Method	Notes
Provisioner	/repo/status/{id}	GET	Status polling (not needed; provision returned synchronously)
Builder	/healthz	GET	Health check
Builder	/images	GET	List images
Builder	/push	POST	Push image to registry