Posit Connect Deployment Guide

In commands below, $SKILL_DIR means the directory containing this file. Replace it with the actual path when executing (e.g., .claude/skills/rsconnect, .cursor/skills/rsconnect).

First-Time Setup: Gitignore

On first use, ensure the agent skills directory is gitignored. The skill folder (e.g., .cursor/, .agents/, .claude/) should not be committed to the project repository.

Check if .gitignore already contains the relevant entry. If not, add it:

# For Cursor-installed skills:
echo ".cursor/" >> .gitignore

# For .agents/ installed skills:
echo ".agents/" >> .gitignore

# For Claude Code skills:
echo ".claude/" >> .gitignore

Rule: Only add the entry for the agent directory that is actually present in the project. Do not add entries for directories that don’t exist.

Before Proceeding

Ask the user if not clear from context:

1. Is this R or Python content?

   • R — Uses renv.lock, rsconnect R package → see R sections below
   • Python — Uses requirements.txt, uv, rsconnect-python → see Python sections below

2. How is this app deployed to Connect?

   • Git-backed — Connect pulls from Git automatically (needs manifest.json) – simplest for teams
   • RStudio IDE — Click “Publish” button in IDE (R only)
   • R console — rsconnect::deployApp() (R only)
   • Jupyter — Publish from notebook interface
   • Command line — rsconnect deploy manifest
   • Not sure — Check Connect UI → Content → Your App → Info tab (shows deployment source)

3. If Git-backed, what branch does Connect watch?

   • Check Connect UI → Content → Your App → Info → “Git Repository” section
   • Recommendation: Use a dedicated deploy branch (e.g., deploy, production) so you control when changes go live

Note: This skill focuses on Git-backed deployment (the default for this project). For push-button methods, the main difference is you don’t need manifest.json — rsconnect generates the bundle on-the-fly.

Windows Notes

• Prefer PowerShell or Git Bash/WSL for command execution.
• If python is not on PATH, use py -3 (Python launcher).
• If Rscript is not on PATH, use Rscript.exe.
• Install uv with PowerShell: powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex".

R Deployment

Quick Start (R)

# 1. Check if ready to deploy
Rscript $SKILL_DIR/scripts/pre_deploy_check.R

# 2. Fix any issues reported, then commit
git add manifest.json renv.lock && git commit -m "chore: update manifest"

# 3. Push (Connect auto-deploys from Git)
git push

First Step: Gather R Project Status

Before doing anything else, run the pre-deploy check to understand the current state:

Rscript $SKILL_DIR/scripts/pre_deploy_check.R

This reports: R version, rsconnect version, manifest status, Source:unknown count, and library sync state. Use the output to inform your response.

R Key Concepts

renv.lock vs manifest.json

They can drift! Always update both together:

renv::snapshot()           # Update renv.lock
rsconnect::writeManifest() # Update manifest.json

What is “Source: unknown”?

Connect needs to know WHERE to download each package. Source: unknown means Connect can’t find it.

Common causes: Package installed from local file, missing repo in options, Bioconductor package without prefix.

Fix: Rscript $SKILL_DIR/scripts/fix_unknown_sources.R --dry-run

R Helper Scripts

# Run from project root
Rscript $SKILL_DIR/scripts/<script>.R

Git Pre-commit Hook

Automatically validate deployment files before each commit:

# Install the pre-commit hook
cat > .git/hooks/pre-commit << 'EOF'
#!/bin/bash
Rscript $SKILL_DIR/scripts/precommit_check.R
EOF
chmod +x .git/hooks/pre-commit

# PowerShell (Windows)
# Replace $SKILL_DIR with your actual skill path
@'
#!/usr/bin/env pwsh
& Rscript $SKILL_DIR/scripts/precommit_check.R
exit $LASTEXITCODE
'@ | Set-Content .git/hooks/pre-commit

Note: Requires PowerShell 7+ (pwsh). If unavailable, run the bash snippet in Git Bash or WSL.

The hook checks:

• No Source: unknown packages in staged renv.lock
• Manifest freshness when renv.lock is staged
• Manifest exists when renv.lock is staged

To bypass (not recommended): git commit --no-verify

R Common Workflows

Deploy After Code Changes (R)

renv::snapshot()  # If dependencies changed
Rscript $SKILL_DIR/scripts/regenerate_manifest.R
git add renv.lock manifest.json
git commit -m "chore: update manifest"
git push

Fix Source:unknown Error

# 1. See what would change (safe)
Rscript $SKILL_DIR/scripts/fix_unknown_sources.R --dry-run

# 2. Apply fixes (creates backup)
Rscript $SKILL_DIR/scripts/fix_unknown_sources.R

# 3. Regenerate manifest
Rscript $SKILL_DIR/scripts/regenerate_manifest.R

Rollback a Bad Deployment

# Option 1: Revert to previous commit
git revert HEAD
git push

# Option 2: Restore from backup (if fix_unknown_sources.R was run)
cp renv.lock.backup.YYYYMMDD_HHMMSS renv.lock
Rscript $SKILL_DIR/scripts/regenerate_manifest.R
git add -A && git commit -m "fix: rollback deployment"
git push

# Option 3: In Connect UI
# Go to Content → Your App → History → Activate previous bundle

R Version Upgrade

1. Install new R version
2. Back up: cp renv.lock renv.lock.bak
3. From new R session:

   renv::init()  # Select option 2: re-initialize
   

4. Fix any Source:unknown issues
5. Regenerate manifest and deploy

See references/troubleshooting.md for detailed migration steps.

Python Deployment (uv)

Quick Start (Python / uv)

# 1. Check if ready to deploy
python $SKILL_DIR/scripts/pre_deploy_check_py.py

# 2. Fix any issues reported, then commit
git add manifest.json requirements.txt && git commit -m "chore: update manifest"

# 3. Push (Connect auto-deploys from Git)
git push

First Step: Gather Python Project Status

Before doing anything else, run the Python pre-deploy check:

python $SKILL_DIR/scripts/pre_deploy_check_py.py

This reports: Python version, uv/rsconnect status, pyproject.toml presence, manifest status, requirements.txt sync, and allow_uv setting. Use the output to inform your response.

Python Key Concepts

requirements.txt vs manifest.json

They can drift! Always update both together:

uv export --no-hashes -o requirements.txt  # Update requirements.txt
rsconnect write-manifest <type> .           # Update manifest.json

uv Workflow

uv is a fast Python package manager. Key commands:

uv export requires pyproject.toml. If it’s missing, regenerate_manifest_py.py can create a minimal one, but review it before committing.

What is allow_uv?

As of Connect 2024.12.0, Connect can use uv pip instead of pip for faster Python package installs. Setting allow_uv: true in manifest.json’s python.package_manager section opts in.

The regenerate_manifest_py.py script patches this automatically. Currently rsconnect-python CLI doesn’t have a flag for it — the script patches manifest.json after generation.

Python Helper Scripts

# Run from project root
python $SKILL_DIR/scripts/<script>.py

Content Type Detection

The regenerate_manifest_py.py script auto-detects your framework: It scans common entrypoints in the project root, src/, and importable package directories (plus pyproject entry points).

Override with: python $SKILL_DIR/scripts/regenerate_manifest_py.py --type fastapi

Python Common Workflows

Deploy After Code Changes (Python)

uv export --no-hashes -o requirements.txt  # If dependencies changed
python $SKILL_DIR/scripts/regenerate_manifest_py.py
git add requirements.txt manifest.json
git commit -m "chore: update manifest"
git push

Regenerate Manifest (Python)

# Auto-detect framework
python $SKILL_DIR/scripts/regenerate_manifest_py.py

# Specify framework explicitly
python $SKILL_DIR/scripts/regenerate_manifest_py.py --type fastapi

# Skip uv export (use existing requirements.txt)
python $SKILL_DIR/scripts/regenerate_manifest_py.py --no-uv-export

Python Version Upgrade

1. Update .python-version with exact major.minor.patch (e.g. 3.13.6, not 3.13) — Posit Connect requires exact version pinning
2. Run: uv sync
3. Export: uv export --no-hashes -o requirements.txt
4. Regenerate manifest: python $SKILL_DIR/scripts/regenerate_manifest_py.py
5. Commit and push

Rollback a Bad Deployment (Python)

# Option 1: Revert to previous commit
git revert HEAD
git push

# Option 2: In Connect UI
# Go to Content → Your App → History → Activate previous bundle

Shared Concepts (R & Python)

Deployment Methods

Git-backed (This Project)

How it works:

1. You generate manifest.json locally
2. You push to Git (with manifest.json + renv.lock/requirements.txt)
3. Connect polls your branch (configurable, default 15 min) and auto-deploys

Why it’s the easiest for teams:

• No credentials needed on dev machines (just Git access)
• Deployment = git push
• Full audit trail in Git history
• Easy rollback (git revert)

Requirements (R): rsconnect ≥ 0.8.15, manifest.json committed to repo Requirements (Python): rsconnect-python, manifest.json committed to repo

Note: Once Git-backed, stay Git-backed. You can’t switch to push-button without recreating the content in Connect.

Branch Strategy

Recommended: Deploy Branch

main (development)
  │
  └── deploy (Connect watches this branch)

Workflow (R):

git checkout deploy
git merge main
Rscript $SKILL_DIR/scripts/regenerate_manifest.R
git add manifest.json renv.lock
git commit -m "chore: update manifest for deploy"
git push origin deploy

Workflow (Python):

git checkout deploy
git merge main
python $SKILL_DIR/scripts/regenerate_manifest_py.py
git add manifest.json requirements.txt
git commit -m "chore: update manifest for deploy"
git push origin deploy

Multi-environment Setup

main (development)
  ├── staging (Connect staging server watches)
  └── production (Connect prod server watches)

Each Connect server points to a different branch. Merge up through environments: main → staging → production

Pre-deploy Checklist

R Projects

• Run Rscript $SKILL_DIR/scripts/pre_deploy_check.R
• No Source:unknown packages
• Manifest is up to date
• .rscignore excludes dev artifacts
• On correct branch for deployment
• Commit and push

Python Projects

• Run python $SKILL_DIR/scripts/pre_deploy_check_py.py
• requirements.txt is up to date
• pyproject.toml exists (required for uv export)
• Manifest is up to date
• allow_uv: true in manifest (recommended)
• .rscignore excludes dev artifacts (.venv/, __pycache__/, etc.)
• On correct branch for deployment
• Commit and push

Additional Resources

• Script docs: scripts/README.md
• Troubleshooting: references/troubleshooting.md
• Commands: references/commands.md

External Links

• Posit Connect Git-backed docs
• renv documentation
• rsconnect writeManifest
• uv documentation
• rsconnect-python docs