Menu
Controlled VM/Computer execution lane for GUI, browser, sandbox, and screenshot-backed tasks.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
Strict parser and validator for Workbench closeout comments, verdict/status discipline, PR reference types, and cross-issue REMAINING sync.
Scout, pilot, or review Mirage virtual-filesystem tool use for Workbench agents while preserving source-of-truth and public-safety boundaries.
Source-first, self-loop resistant guardrails for Capy GitHub dialogue responders before any write-capable PR, issue, or review action.
Disk, swap, VM, agent workspace, and stale session hygiene for Multica workbench runtimes.
Advisory-only algorithm and data-structure review gate between Technical Design and Task List.
Durable memory, session summaries, decision logs, issue closeout synthesis, and compact handoffs.
| name | workbench-capy-vm-lane |
| description | Controlled VM/Computer execution lane for GUI, browser, sandbox, and screenshot-backed tasks. |
Use this skill only when an issue declares EXECUTION_TARGET: capy-vm or asks for isolated GUI/browser/sandbox execution.
Before execution:
VM_LEASE.secrets_policy.network_policy.ARTIFACTS_REQUIRED.TEARDOWN.If any field is missing, return BLOCKED with the smallest missing field list.
secrets_policy: nonenetwork_policy: localhost-only for smoke testsTEARDOWN: destroyCAPY_VM_ARTIFACT_DIR explicitly for durable review evidenceThe local smoke path uses a Scrapybara-compatible Computer API:
curl -fsS -X POST "http://127.0.0.1:${CAPY_API_PORT}/computer" \
-H "Content-Type: application/json" \
-d '{"action":"take_screenshot"}'
The response must be saved before summarizing.
Always report:
VERDICT: PASS / FLAG / BLOCK.Never paste base64 screenshots, raw cookies, OAuth codes, private tokens, API keys, or unrelated screen content into comments.