Menu
Connects any Claude Code project to the AgenticOS Universal Hub — enables multi-project monitoring, iPhone approval, Ollama handoff, and skill auto-loading.
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
Wraps gsd-pi (Get Shit Done coding agent CLI) with Marcus's project constraints and coding standards so gsd tasks inherit observer pattern, no em-dashes, no Co-Authored-By, and AgenticOS event integration automatically. Use when the user says "run this with gsd", "hand this off to gsd", "use gsd-pi for this", "delegate to gsd", or wants to execute a coding task using gsd-pi as the execution engine rather than Claude Code directly. Also trigger when autonomous-workflow needs to fan out a phase to gsd-pi for execution.
Meta-skill that runs the full skill lifecycle in one command: create, eval, improve, and publish. Use when the user says "forge a skill", "create and improve a skill", "run the skill loop on X", "build and publish a skill", "skill-forge", or wants to go from idea to production-ready skill without manually chaining skill-creator and skill-self-improver. Also trigger when the user wants to combine or upgrade existing skills, write eval.json for a skill, or push skills to GitHub. Wraps skill-creator + skill-self-improver into one /skill-forge command.
Recover from interrupted AI agent coding sessions by detecting the exact interruption point, reconstructing intent from pre-written step comments, completing unfinished implementations, and validating through build + supervisor review. Use this skill whenever a user mentions session recovery, interrupted work, picking up where an agent left off, completing unfinished code, finding where a session dropped, resuming development, or anything related to detecting and finishing incomplete AI-assisted coding sessions. Also trigger when users mention Claude Code agent sessions that were cut off, lost progress, or need continuation. This skill works across all languages and frameworks — C++, C#, Python, TypeScript, Rust, Java, Go, Unreal Engine, Unity, .NET, React, Node.js. Trigger aggressively — if a user says anything about resuming, continuing, recovering, or picking up work, use this skill.
Agent state-writing protocol for the AgenticOS Command Center. Any Claude sub-agent running in a multi-agent parallel session reads this skill to participate in dashboard tracking, approval gates, and reviewer spawning.
Cross-domain protocol for supervised parallel multi-agent runs against the AgenticOS Command Center. Use whenever Marcus says "spawn parallel agents", "multi-agent task", "review agent output", "approval gate", "monitor sub-agents", "supervised pipeline", "gate the next stage", "run several agents", "watch the agents", or kicks off a session with named AGENT-01, AGENT-02 roster. Defines the AgentState contract, the approval gate handshake, the reviewer-agent pattern, and the parallel etiquette every sub-agent must follow to appear in the Command Center dashboard with no extra wiring.
Workflow automation and multi-agent orchestration tools. Use when the user mentions n8n, Langflow, claude-squad, container-use, Dagger, workflow automation, pipeline orchestration, multi-agent coordination, or wants to set up automated workflows. Also trigger for 'automate this process', 'create a workflow', 'run agents in parallel', or integration of multiple tools into a pipeline.
| name | agentic-hub |
| description | Connects any Claude Code project to the AgenticOS Universal Hub — enables multi-project monitoring, iPhone approval, Ollama handoff, and skill auto-loading. |
| version | 1.0.0 |
| author | Marcus Daley |
Auto-connects a new Claude Code project to the AgenticOS Universal Hub running at C:/ClaudeSkills/AgenticOS. Once connected, Marcus can:
curl -X POST http://localhost:7842/projects/register \
-H "Content-Type: application/json" \
-d '{"path": "'$(pwd)'"}'
Or just create/edit CLAUDE.md — the project_watcher daemon registers it automatically within 5 seconds.
After each work session, record what Marcus needs to do next:
curl -X POST http://localhost:7842/projects/<PROJECT_ID>/phase \
-H "Content-Type: application/json" \
-d '{"hint": "Review the auth system changes in src/api/ then run tests"}'
When Claude Code is approaching its context limit, call:
curl -X POST http://localhost:7842/handoff/snapshot \
-H "Content-Type: application/json" \
-d '{
"project_name": "MyProject",
"project_path": "C:/Users/daley/Projects/MyProject",
"plan_summary": "...",
"completed_tasks": ["task 1", "task 2"],
"pending_tasks": ["task 3", "task 4"],
"context_notes": "...",
"files_modified": ["src/api/auth.py"],
"next_action": "Implement the token refresh endpoint"
}'
Ollama will pick up the manifest and continue automatically.
Add skill references to your CLAUDE.md like:
skills/universal-coding-standards/
skills/architecture-patterns/
The skill_loader will copy them into .claude/skills/ automatically.
Claude Code writes manifest → Ollama continues work → Claude Code reviews
Start the runner manually:
python -m AgenticOS.handoff_runner
Or set OLLAMA_HANDOFF_MODEL in .env to configure which model runs.
| Endpoint | Method | Purpose |
|---|---|---|
| /projects | GET | List all projects |
| /projects/active | GET | List active projects |
| /projects/register | POST | Register a project |
| /projects/{id}/phase | GET | What to do now |
| /projects/{id}/phase | POST | Set phase hint |
| /handoff | GET | Handoff manifest status |
| /handoff/snapshot | POST | Write handoff manifest |
| /agents | GET | All running agents |
| /approve/{id} | POST | Approve agent |
| /research/{id} | POST | Request more research |
| /review/{id} | POST | Trigger reviewer |