| name | octoarms-hydra-model-swarm |
| description | Coordinate OctoArmor, HYDRA swarm extensions, Hugging Face model deployment, and governed multi-agent task completion across browser, deep-research, Colab bridge, universal synthesis, and Spiralverse intent-auth lanes. |
Octo Arms Hydra Model Swarm
Use this skill when a task needs one governed control layer over:
- OctoArmor model routing
- HYDRA browser or research swarms
- Hugging Face model selection, deployment, or push lanes
- Colab bridge handoff into automation or training
- Sacred Tongues routing and Spiralverse intent-auth
- reproducible action maps, evidence packets, and return contracts
This is an orchestration skill. It should compose the existing skills named below rather than replacing them.
Trigger Conditions
Use this skill when the user asks to:
- launch or coordinate a multi-agent HYDRA or OctoArmor swarm
- route tasks across local models, Hugging Face models, and browser agents
- deploy or test Hugging Face-backed agent lanes for real task completion
- connect Colab compute into governed agent workflows
- run one task through multiple SCBE skills with a single command plan
- enforce Sacred Tongues or Spiralverse intent-auth before execution
- turn a large task into bounded arms, packets, and return contracts
Skill Composition
Load these skills in this order unless the task is narrower:
$scbe-universal-synthesis for skill inventory and Sacred Tongues routing.
$scbe-spiralverse-intent-auth for ALLOW, QUARANTINE, or DENY packet logic.
$hydra-node-terminal-browsing when deterministic browse evidence is needed.
$hydra-clawbot-synthesis when the workflow crosses browser, dataset, Airtable, HF, or Obsidian lanes.
$hydra-deep-research-self-healing when the job should run continuously or recover automatically.
$scbe-n8n-colab-bridge when a Colab runtime or local notebook bridge is part of the execution path.
Read skill-composition.md when you need the lane-by-lane handoff contract.
Read repo-surfaces.md when you need the concrete repo commands, files, and services.
Fast Start
Use the bundled helper when you want packetize + dispatch in one call.
python C:\Users\issda\.codex\skills\octoarms-hydra-model-swarm\scripts\octoarms_dispatch.py --repo-root C:/Users/issda/SCBE-AETHERMOORE --task "route a governed multi-agent task" --lane octoarmor-triage
python C:\Users\issda\.codex\skills\octoarms-hydra-model-swarm\scripts\octoarms_dispatch.py --repo-root C:/Users/issda/SCBE-AETHERMOORE --task "research governed browser agents" --lane hydra-swarm --dry-run
The helper always:
- generates a flow plan
- packetizes the task into bounded arms
- applies a low-cost OctoArmor routing overlay
- optionally launches the selected lane
Core Workflow
1. Establish the control plane
Start from the repo root and verify the operator shell is live.
cd C:\Users\issda\SCBE-AETHERMOORE
.\scripts\install_hydra_quick_aliases.ps1
issac-help
hstatus
hqueue
If OctoArmor or HYDRA surfaces are missing, use the command-center aliases in repo-surfaces.md.
2. Choose the execution lane
Use the smallest lane that can finish the task.
- OctoArmor routing only:
- inspect
src/aetherbrowser/router.py
- use
octo-serve for the model gateway
- HYDRA browser swarm:
- use
python -m hydra.cli_swarm --dry-run ... first
- HYDRA terminal evidence:
- use the Node browse script for deterministic JSON
- Deep research:
- use the self-healing loop only for multi-cycle or continuous jobs
- Colab or training bridge:
- use
scbe.py colab ... or the Colab bridge skill before execution
- Hugging Face model lane:
- pass a specific model id when using
--provider hf
Default preference:
- low-complexity or cheap checks -> local
- bounded remote specialist run -> Hugging Face
- browser-heavy evidence gathering -> HYDRA browse lanes
- long-form resilient research -> self-healing lane
3. Apply governed intent routing
Before live execution, produce an intent packet and gate decision.
Minimum contract:
- choose a tongue or tongue sequence
- state task intent
- define recipients or runtime lane
- set a return artifact path
If the governance decision is not ALLOW, do not execute the live action. Fall back to dry-run, planning, or a reduced-scope packet.
4. Execute in bounded arms
Treat each arm as a bounded worker packet with:
task_id
objective
lane
provider
dependencies
artifacts_out
return_contract
Good examples:
- Arm 1: route with OctoArmor, return selected provider snapshot
- Arm 2: run deterministic browse evidence on target URLs
- Arm 3: run HYDRA swarm with
--dry-run, then live if clean
- Arm 4: sync approved outputs to HF or Colab bridge
- Arm 5: emit action-map and cross-talk packet
5. Capture proof
Persist evidence whenever the task matters.
Recommended outputs:
- browse JSON artifact
- swarm result JSON
- action-map run id
- Colab bridge status
- model id and provider used
- final governance decision packet
6. Close the loop
Return:
- what lane ran
- what model/provider ran
- whether intent-auth allowed it
- where artifacts were saved
- what the next bounded action is
Quick Commands
OctoArmor and HYDRA
octo-serve
python -m hydra.cli_swarm --status
python -m hydra.cli_swarm --dry-run "research governed browser agents"
python -m hydra.cli_swarm --provider hf --model HuggingFaceTB/SmolLM2-1.7B-Instruct "summarize target system"
Deterministic browse evidence
node C:\Users\issda\.codex\skills\hydra-node-terminal-browsing\scripts\hydra_terminal_browse.mjs --url "https://example.com" --out "artifacts\page_evidence.json"
Deep-research recovery loop
powershell -NoProfile -ExecutionPolicy Bypass -File C:\Users\issda\SCBE-AETHERMOORE\scripts\system\run_deep_research_self_healing.ps1 -Topic "governed model swarm" -Continuous -SleepSeconds 120 -UsePlaywriter -RunCiTriage
Colab bridge
python scbe.py colab bridge-status --name pivot
python scbe.py colab bridge-set --name pivot --backend-url "http://127.0.0.1:8888/?token=..." --probe
python scbe.py colab bridge-probe --name pivot
Universal synthesis refresh
python C:/Users/issda/SCBE-AETHERMOORE/scripts/system/refresh_universal_skill_synthesis.py
Operational Rules
- Prefer
--dry-run before any live browser swarm action.
- Prefer the cheapest valid provider first; escalate only when the task needs it.
- Never hardcode tokens, model secrets, or Colab tokens into skill files or output artifacts.
- Do not execute live actions when intent-auth returns
QUARANTINE or DENY.
- Keep work packets bounded. If a job is too big for one arm, packetize it instead of improvising.
- Record the selected model, provider, lane, and artifact paths for every meaningful run.
- Use deep-research self-healing only for jobs that truly need recovery logic.
References
Resources
scripts/
scripts/octoarms_dispatch.py
Packetizes a task with scbe-system flow plan and flow packetize, scores it with OctoArmor, then launches one selected lane: triage, HYDRA swarm, browse evidence, or Colab bridge status/probe.