Skill deployer

Skill Deployer

Deploy skills from the sandbox filesystem into a hosted project UI - making them visible in the project sidebar, available as slash commands, and persistent across tasks. This is a compatibility adapter; universal cross-AI setups should use skill-router skill <name> and OpenSkills-style sync instead.

How It Works

Skills in /home/ubuntu/skills/ are invisible in many hosted project UIs by default. This skill bridges that gap:

1. Packages each skill directory into a .zip (SKILL.md + scripts/references at root)
2. Authenticates with the provider gRPC-web API using a JWT from the browser session
3. Uploads zips to specific projects via UploadProjectSkill endpoint
4. Skills then appear in the project sidebar under "Skills" and as slash commands in new tasks

Prerequisites

Before deploying, two things are needed:

1. JWT Token — Extract from the browser while the user is logged into the target provider web app:

// Run via browser_console_exec tool:
(document.cookie.split(';').find(c => c.trim().startsWith('session_id=')) || '').split('=').slice(1).join('=') || 'TOKEN_NOT_FOUND'

Save to file: echo "TOKEN" > /tmp/skill_deployer_token.txt

2. Project UIDs — Found in the URL when viewing a project:

/app/project/<PROJECT_UID>

Extract from sidebar via browser_console_exec or by clicking each project.

Commands

Package Skills (No Upload)

Create zip files from all skills in the skills directory:

python3 /home/ubuntu/skills/skill-deployer/scripts/deploy_skills.py \
  --token "$(cat /tmp/skill_deployer_token.txt)" --package

List Skills in a Project

See what's currently in a project's UI:

python3 /home/ubuntu/skills/skill-deployer/scripts/deploy_skills.py \
  --token "$(cat /tmp/skill_deployer_token.txt)" --project-uid UID --list

Upload a Single Skill to UI

Package and upload one skill by name:

python3 /home/ubuntu/skills/skill-deployer/scripts/deploy_skills.py \
  --token "$(cat /tmp/skill_deployer_token.txt)" --project-uid UID --upload-skill skill-name

Upload All Skills from Zip Directory

python3 /home/ubuntu/skills/skill-deployer/scripts/deploy_skills.py \
  --token "$(cat /tmp/skill_deployer_token.txt)" --project-uid UID --upload-dir /tmp/skill_zips

Delete a Skill from UI

python3 /home/ubuntu/skills/skill-deployer/scripts/deploy_skills.py \
  --token "$(cat /tmp/skill_deployer_token.txt)" --project-uid UID --delete-name skill-name

Sync Project to Exact Skill Set

Deletes skills not in the list, uploads missing ones:

python3 /home/ubuntu/skills/skill-deployer/scripts/deploy_skills.py \
  --token "$(cat /tmp/skill_deployer_token.txt)" --project-uid UID \
  --sync skill-a skill-b skill-c

Multi-Project Deployment with LLM Curation

For deploying curated, domain-relevant skills to multiple projects at once.

Step A — Create projects file (projects.json):

{
  "projects": [
    {
      "name": "My Web App",
      "uid": "abc123",
      "context": "Full-stack React + Node web app with auth and database",
      "tasks": ["frontend dev", "API design", "database queries"],
      "max_skills": 300
    }
  ]
}

Step B — Run LLM curation (selects universal + domain-specific skills):

python3 /home/ubuntu/skills/skill-deployer/scripts/curate_for_project.py \
  --projects projects.json --output deployment_plan.json

Step C — Preview the plan:

python3 /home/ubuntu/skills/skill-deployer/scripts/deploy_skills.py \
  --token "$(cat /tmp/skill_deployer_token.txt)" --plan deployment_plan.json --dry-run

Step D — Deploy to all projects:

python3 /home/ubuntu/skills/skill-deployer/scripts/deploy_skills.py \
  --token "$(cat /tmp/skill_deployer_token.txt)" --plan deployment_plan.json

Options

Flag	Default	Description
--api-base	none	Provider API base URL; required for API calls unless SKILL_DEPLOYER_API_BASE is set
--skills-dir	/home/ubuntu/skills	Source skills directory
--zip-dir	/tmp/skill_zips	Directory for packaged zips
--rate-limit	0.15	Seconds between API calls
--max-retries	3	Retries on 429 rate limit errors
--dry-run	off	Preview changes without executing

Limits and Constraints

• 500 skills per project — common provider-enforced hard limit
• Rate limiting — 429 errors handled with exponential backoff
• Token expiry — JWT tokens last ~24 hours; re-extract if 401 errors occur
• SKILL.md required — Zips must contain valid SKILL.md with YAML frontmatter at root

Troubleshooting

Error	Fix
401 Unauthorized	Token expired. Re-extract from browser.
Missing API base	Pass --api-base https://provider.example or set SKILL_DEPLOYER_API_BASE.
429 Rate Limited	Increase --rate-limit to 0.3+.
500 SKILL_MD_INVALID_FORMAT	Fix SKILL.md YAML frontmatter in the source skill.
Project limit reached	Max 500. Use curation to select only the best skills.
Skills not showing in UI	Refresh the project page in browser. Skills appear in sidebar.

API Reference

Provider endpoint details vary by hosted UI. Use the provider's gRPC-web project-skill API reference when setting --api-base.