multi-agent-cloud-offload
Deterministically sort, bundle, verify, and offshore local files through multiple AI/model lanes while capturing training rows and method evidence. Use when Codex needs to inventory folders, batch-process files, upload them to cloud targets such as rclone-backed Google Drive, Hugging Face, or GitHub, and only delete sources after the configured number of verified targets succeed.
Best use case
multi-agent-cloud-offload is best used when you need a repeatable AI agent workflow instead of a one-off prompt.
Deterministically sort, bundle, verify, and offshore local files through multiple AI/model lanes while capturing training rows and method evidence. Use when Codex needs to inventory folders, batch-process files, upload them to cloud targets such as rclone-backed Google Drive, Hugging Face, or GitHub, and only delete sources after the configured number of verified targets succeed.
Teams using multi-agent-cloud-offload should expect a more consistent output, faster repeated execution, less prompt rewriting.
When to use this skill
- You want a reusable workflow that can be run more than once with consistent structure.
When not to use this skill
- You only need a quick one-off answer and do not need a reusable workflow.
- You cannot install or maintain the underlying files, dependencies, or repository context.
Installation
Claude Code / Cursor / Codex
Manual Installation
- Download SKILL.md from GitHub
- Place it in
.claude/skills/multi-agent-cloud-offload/SKILL.mdinside your project - Restart your AI agent — it will auto-discover the skill
How multi-agent-cloud-offload Compares
| Feature / Agent | multi-agent-cloud-offload | Standard Approach |
|---|---|---|
| Platform Support | Not specified | Limited / Varies |
| Context Awareness | High | Baseline |
| Installation Complexity | Unknown | N/A |
Frequently Asked Questions
What does this skill do?
Deterministically sort, bundle, verify, and offshore local files through multiple AI/model lanes while capturing training rows and method evidence. Use when Codex needs to inventory folders, batch-process files, upload them to cloud targets such as rclone-backed Google Drive, Hugging Face, or GitHub, and only delete sources after the configured number of verified targets succeed.
Where can I find the source code?
You can find the source code on GitHub using the link provided at the top of the page.
SKILL.md Source
# Multi Agent Cloud Offload Use this skill when file loss would be unacceptable and the task requires a strict inventory-to-verification pipeline rather than ad hoc copying. ## Core Workflow 1. Read the control files first: - `scripts/multi_agent_offload.py` - `scripts/multi_agent_offload.json` - `scripts/run_multi_agent_offload.ps1` 2. Treat source deletion as blocked until destination verification succeeds. 3. Prefer `rclone` targets with remote hash verification for delete authority. 4. Keep `required_verified_targets` at `2` or higher unless the user explicitly accepts lower durability. 5. Run a dry-run before any live offload: ```powershell pwsh -File scripts/run_multi_agent_offload.ps1 -DryRun -NoProcess -MaxFiles 3 ``` 6. Run a live non-deleting pass before any delete pass: ```powershell pwsh -File scripts/run_multi_agent_offload.ps1 -MaxFiles 3 ``` 7. Use `-DeleteSource` only after the run artifacts show the required number of verified targets for each file: ```powershell pwsh -File scripts/run_multi_agent_offload.ps1 -MaxFiles 3 -DeleteSource ``` ## Verification Rule - Never delete after a plain copy or API acknowledgement alone. - Delete only when the run output proves the configured verification gate passed. - The default runner writes per-file evidence into: - `training/runs/multi_agent_offload/<run_id>/file_results.json` - `training/runs/multi_agent_offload/<run_id>/run_summary.json` - `training/runs/multi_agent_offload/<run_id>/method_registry.json` ## Training Capture Every processed file should produce a training/example row at: - `training/runs/multi_agent_offload/<run_id>/training_rows.jsonl` Use those rows to curate later uploads to Hugging Face datasets. Treat them as audit artifacts first and training data second. ## Adjustment Points - Change source folders, lane models, and shipping targets in `scripts/multi_agent_offload.json`. - Use `--source-root` overrides for narrow test passes. - Use `--targets` to constrain a run to specific destinations. - Use `--reprocess` when the routing policy or model mix changed and old successes are no longer authoritative. ## Reference Read `references/runbook.md` when changing durability policy, target ordering, or the live run procedure.
Related Skills
cloud-storage-local-storage-management
Audit, sort, route, and back up large Windows workspaces across local disks and cloud-sync folders without recursive mirror failures. Use when the user wants the filesystem cleaned up, cloud storage organized, backup targets chosen by free space and route quality, or hardware/storage bottlenecks identified before moving data.
multi-model-animation-studio-notes
Organize animation and manhwa/webtoon production notes across multiple AI models and reference sources. Use when extracting reference frames from recap videos or webtoon pages, planning scene packets, comparing model lanes, or preserving arc-level world continuity while allowing deliberate per-panel style shifts.
multi-agent-review-gate
Run a structured review gate before merging multi-agent outputs. Use when multiple agents have produced work packets that need quality checks, conflict detection, and approval before integration.
multi-agent-orchestrator
Coordinate multi-agent workflows end to end. Use when a task should be decomposed into parallel sub-work with explicit ownership, dependency ordering, collision avoidance, checkpoint tracking, and final output recomposition.
scbe-training-pair-authoring
Create prompt and response and metadata training pairs from SCBE documents, repair traces, terminal sessions, and operational workflows using the repository's canonical dataset contract and provenance rules.
scbe-spin-conversation-engine
Generate SFT training data via radial matrix conversation pivots with D&D-style combat research mode. Produces diverse, cost-effective training pairs with Sacred Tongue encoding, golden spiral problem distribution, and harmonic re-attunement.
scbe-research-training-bridge
Stage arXiv evidence and Obsidian markdown into source-grounded Hugging Face training bundles for research, review, and later SFT runs.
scbe-document-management
Consolidate overlapping docs, classify files by authority, and keep SCBE repo documents aligned with runtime truth. Use when the repo has drift between canonical docs, public docs, proposal notes, research branches, and generated evidence.
scbe-colab-bridge
Control Google Colab notebooks from Claude Code via Chrome extension. Execute cells, run terminal commands, read outputs, and manage GPU compute remotely.
scbe-claim-to-code-evidence
Map SCBE Notion technical claims, proof pages, and patent-facing architecture notes to concrete repository evidence such as code paths, tests, demos, and docs. Use when Codex needs to build a due-diligence packet, claim-to-code audit, implementation crosswalk, patent support note, or proof summary from local Notion exports and repo artifacts.
scbe-autonomous-worker-productizer
Turn SCBE automation, autonomous worker, and revenue-system notes into concrete offers, workflow packs, pilot plans, or SaaS-facing product packets. Use when Codex needs to package Notion automation pages into buyer-ready offerings, n8n/Zapier workflow designs, flock-backed worker systems, or implementation roadmaps tied to existing SCBE repo surfaces.
long-form-work-orchestrator
Run long-form engineering work in checkpointed phases with deterministic artifacts, resilience handling, and end-of-run reliability reporting.