{"id":74304,"plugin_id":"plugins_6ac4ce7367e88191bdf6378dc7d46962","kind":"skill","collection_source":"plugin_package","comparison_source":null,"observed_at":"2026-10-31T06:11:15.725Z","digest":"397b11120ef721d705b39efe5eb0fbdc8ba9b6b94dbcd438468be53e7cda6dbf","against":null,"payload":{"description":"Use when a large build has been decomposed into a multi-slice plan (contracts, acceptance tests, dependency DAG) so the codex headless executor implements each slice while Codex orchestrates and judges.","included_files":[{"relative_path":".gitignore","size_in_bytes":39},{"relative_path":"LICENSE","size_in_bytes":1088},{"relative_path":"PRIVACY.md","size_in_bytes":8943},{"relative_path":"README.md","size_in_bytes":4127},{"relative_path":"SECURITY.md","size_in_bytes":3623},{"relative_path":"agents/openai.yaml","size_in_bytes":653},{"relative_path":"examples/demo-plan.md","size_in_bytes":508},{"relative_path":"references/architecture.md","size_in_bytes":3078},{"relative_path":"references/authoring-plans.md","size_in_bytes":7716},{"relative_path":"references/codex-workflow.md","size_in_bytes":3522},{"relative_path":"references/delivery-core.md","size_in_bytes":11007},{"relative_path":"references/observability.md","size_in_bytes":3867},{"relative_path":"references/provider-setup.md","size_in_bytes":6157},{"relative_path":"references/provider.md","size_in_bytes":2785},{"relative_path":"scripts/cld/__init__.py","size_in_bytes":0},{"relative_path":"scripts/cld/__main__.py","size_in_bytes":283},{"relative_path":"scripts/cld/_skeleton.py","size_in_bytes":46},{"relative_path":"scripts/cld/_windows_job.py","size_in_bytes":3126},{"relative_path":"scripts/cld/_windows_workspace.py","size_in_bytes":2791},{"relative_path":"scripts/cld/acceptance.py","size_in_bytes":7015},{"relative_path":"scripts/cld/accounting.py","size_in_bytes":14338},{"relative_path":"scripts/cld/admission.py","size_in_bytes":6797},{"relative_path":"scripts/cld/attempts.py","size_in_bytes":4303},{"relative_path":"scripts/cld/behavioral.py","size_in_bytes":2509},{"relative_path":"scripts/cld/build_state.py","size_in_bytes":8782},{"relative_path":"scripts/cld/candidate.py","size_in_bytes":12399},{"relative_path":"scripts/cld/cli.py","size_in_bytes":72161},{"relative_path":"scripts/cld/cli_response.py","size_in_bytes":5690},{"relative_path":"scripts/cld/dag.py","size_in_bytes":2085},{"relative_path":"scripts/cld/evidence.py","size_in_bytes":3018},{"relative_path":"scripts/cld/executors/__init__.py","size_in_bytes":340},{"relative_path":"scripts/cld/executors/_capture.py","size_in_bytes":3292},{"relative_path":"scripts/cld/executors/base.py","size_in_bytes":3281},{"relative_path":"scripts/cld/gc.py","size_in_bytes":9591},{"relative_path":"scripts/cld/hook.py","size_in_bytes":2012},{"relative_path":"scripts/cld/integration.py","size_in_bytes":12656},{"relative_path":"scripts/cld/integration_gate.py","size_in_bytes":1018},{"relative_path":"scripts/cld/judge.py","size_in_bytes":3429},{"relative_path":"scripts/cld/ledger.py","size_in_bytes":9189},{"relative_path":"scripts/cld/locking.py","size_in_bytes":1454},{"relative_path":"scripts/cld/models.py","size_in_bytes":34138},{"relative_path":"scripts/cld/native_cli.py","size_in_bytes":3259},{"relative_path":"scripts/cld/orchestrator.py","size_in_bytes":40162},{"relative_path":"scripts/cld/picker.py","size_in_bytes":2320},{"relative_path":"scripts/cld/plan/__init__.py","size_in_bytes":0},{"relative_path":"scripts/cld/plan/slice.py","size_in_bytes":5463},{"relative_path":"scripts/cld/process.py","size_in_bytes":12672},{"relative_path":"scripts/cld/providers_api.py","size_in_bytes":6641},{"relative_path":"scripts/cld/recovery.py","size_in_bytes":12879},{"relative_path":"scripts/cld/repair.py","size_in_bytes":6398},{"relative_path":"scripts/cld/status.py","size_in_bytes":10471},{"relative_path":"scripts/cld/summary.py","size_in_bytes":5817},{"relative_path":"scripts/cld/telemetry.py","size_in_bytes":7982},{"relative_path":"scripts/cld/test_run.py","size_in_bytes":2585},{"relative_path":"scripts/cld/usage.py","size_in_bytes":4826},{"relative_path":"scripts/cld/validate.py","size_in_bytes":10635},{"relative_path":"scripts/cld/worktree.py","size_in_bytes":5149},{"relative_path":"scripts/cld_providers/__init__.py","size_in_bytes":71},{"relative_path":"scripts/cld_providers/codex/SKILL.fragment.md","size_in_bytes":2785},{"relative_path":"scripts/cld_providers/codex/__init__.py","size_in_bytes":115},{"relative_path":"scripts/cld_providers/codex/catalog.py","size_in_bytes":2926},{"relative_path":"scripts/cld_providers/codex/config.py","size_in_bytes":4301},{"relative_path":"scripts/cld_providers/codex/contract.py","size_in_bytes":8705},{"relative_path":"scripts/cld_providers/codex/launcher.py","size_in_bytes":1711},{"relative_path":"scripts/cld_providers/codex/provider.py","size_in_bytes":11019},{"relative_path":"scripts/cld_providers/codex/setup.md","size_in_bytes":6157},{"relative_path":"scripts/list_models.py","size_in_bytes":448},{"relative_path":"scripts/run_delivery.py","size_in_bytes":5238}],"name":"cross-llm-codex","skill_md_contents":"---\nname: cross-llm-codex\ndescription: >-\n  Use when a large build has been decomposed into a multi-slice plan\n  (contracts, acceptance tests, dependency DAG) so the codex\n  headless executor implements each slice while Codex orchestrates and judges.\n---\n<!-- GENERATED from cross-llm-delivery (provider: codex, v0.4.2) - do not edit here; edit the monorepo source. -->\n\n# Cross-LLM Delivery -- Codex host, codex executor\n\n**Codex is the lead.** You decompose the build into testable slices, fix the\ncontracts, author the acceptance tests, judge every result, and integrate\nverified work. The **codex headless executor** does the bulk\nimplementation typing inside isolated\ngit worktrees. The vendored `cld` engine and driver under `scripts/` run\nin place -- no package install and no source checkout are needed.\nAn exact model ID is required; this provider has no default model. Pass `--executor codex:<model-id>@<effort>` explicitly. For requested fast mode, use `codex:gpt-6-luna@max+fast`; do not silently drop the tier.\n\nFor executor choices, run `python scripts/list_models.py --json` and follow\nthe picker section of `references/delivery-core.md`. Present exact returned\nIDs and supported efforts; discovery does not authorize dispatch.\n\n## When to use (and when not to)\n\n- **Use it** on a large build expressed as a plan of independently testable\n  slices with a dependency DAG, where executor tokens dwarf lead tokens.\n- **Do not use it** for small one-file fixes; the per-dispatch overhead only\n  pays off on large builds. Small work stays with Codex directly.\n\n## Ground rules\n\n- Respect the user's existing project instructions (AGENTS.md and friends).\n  Never install, overwrite, or replace them.\n- Preserve the user's exact executor/model/effort/tier selection. Present the\n  shortlist only when no selection exists; a listing is not dispatch permission.\n  Describe discovery only for the recorded host version and surface.\n- Never claim the provider's cost is free. Check that billed dispatches are\n  covered by the user's existing authorization; ask only if that is unclear.\n- For gate-4 repairs, follow the user's authorization and the host's active\n  approval policy. Never silently dispatch another paid attempt.\n- `--step` and validation make paid network calls through the executor CLI.\n  Check the active host permissions before dispatch. If network is disabled,\n  use the host's supported authorization flow; do not bypass its restrictions.\n  CLD blocks dispatch when `CODEX_SANDBOX_NETWORK_DISABLED=1` is set.\n\n## Drive the build\n\nRun these commands from the installed skill directory, using absolute plan and\ntarget-repo paths. The driver is vendored in this bundle. Prefer `--json` for\nmachine-checkable gates and act on the JSON `gate_code`. On Windows a shell\nwrapper (`pwsh -Command`) can report exit 1 for any non-zero driver exit, so\nthe JSON gate, not the shell's exit status, is authoritative.\n\n```bash\npython scripts/run_delivery.py <plan.md> --repo <dir> --dry-run --json  # preview layers\npython scripts/run_delivery.py <plan.md> --repo <dir> --step --workers 1 --executor <exact-spec> --validation-policy <deny|unmetered|allow> --budget-attempts <N> --json\npython scripts/run_delivery.py --status --repo <dir> --json             # cheap digest\npython scripts/run_delivery.py <plan.md> --repo <dir> --integrate --integration-tests <selector> --json\n```\n\nChoose the validation policy and cumulative attempt limit under the user's\nauthorization. Validation and retries count as dispatches. Prefer one worker\nfor a bounded sitting; the CLI default is four.\n\nExit codes: 0 ok, work remains; 2 failure/defer; 3 integrated and verified;\n4 lead repair required; 5 invalid plan/state or missing prerequisite;\n6 accepted work awaits integration.\n\n## References\n\n- `references/delivery-core.md` -- the shared, host-neutral gate (exit-code)\n  contract and authorization rules; identical in every host bundle.\n- `references/codex-workflow.md` -- the full Codex-led workflow: plan authoring,\n  batch-stepping, the model picker, gates, and the repair loop.\n- `references/provider-setup.md` -- install and verify the codex CLI.\n- `references/provider.md` -- codex executor specifics: locked\n  invocation, auth, cost, and platform notes.\n- [README.md](README.md) -- data handling, privacy, security reporting and a\n  no-inference reviewer preview of this bundle.\n"},"changes":[],"summary":"First saved snapshot. No earlier version is available for comparison.","summary_kind":"deterministic","summary_metadata":{}}