webmaxru/agent-skill-deploy

Agent Skill Collection Deploy

Purpose

Automate multi-surface deployment of agent skill collections:

• Pre-flight validation of git state, skills inventory, and surface readiness
• Conventional commit analysis with version bump recommendation
• Version bumping across all detected surface configuration files
• Surface-specific deployment with dry-run capability
• User approval gates before irreversible operations

When to Use This Skill

Use this skill when the user:

• Asks to "release", "deploy", "publish", or "ship" a skills collection
• Wants to bump versions and push to one or more marketplaces
• Needs to create a GitHub release for a skills repository
• Wants to publish skills to Claude Code, VS Code, or Copilot CLI marketplaces
• Asks to check deployment readiness or run a dry-run release

Do not use for npm packages, Docker deployments, Azure resource provisioning, or repositories without a /skills directory.

Supported Surfaces

| Surface | Config Files | Deploy Action | Tool Required | | ---------------- | ------------------------------------------------------------------ | -------------------------------- | ------------------- | | github | Git remote URL | Create tag + GitHub release | gh | | claude-code | .claude-plugin/plugin.json (required), .claude-plugin/marketplace.json (optional) | Bump plugin version, commit, push | git | | vscode | package.json | Bump version, commit, push | git | | copilot-cli | package.json, .github/plugin/plugin.json, .github/plugin/marketplace.json (optional) | Bump plugin version, commit, push | git |

Bundled Scripts

This skill includes three Node.js helper scripts in scripts/ for cross-platform operation (Windows and macOS):

1. deploy-preflight.mjs — Validates git state, discovers surfaces, checks tool availability, verifies version consistency (including git tag version)
2. deploy-analyze.mjs — Inventories skills, analyzes commits since last tag, recommends version bump
3. deploy-execute.mjs — Bumps versions in all detected config files (regardless of selected surfaces), commits, tags, and deploys to selected surfaces

Run scripts from the skill directory:

node scripts/deploy-preflight.mjs
node scripts/deploy-analyze.mjs
node scripts/deploy-execute.mjs 1.2.0 --surfaces github,claude-code --dry-run

Version Handling Rules

Marketplace.json files contain two distinct version fields with different semantics:

• plugins[].version (plugin version) — Tracks the version of each listed plugin. Bumped during releases for local plugins (where source is ".").
• metadata.version (marketplace version) — Tracks the version of the marketplace collection itself. Never bumped during skill/plugin releases. Managed independently.

Cross-file sync rules:

• All plugin versions must stay consistent: plugin.json, package.json, plugins[].version in marketplace files, and the git tag.
• All metadata.version values across marketplace.json files (.claude-plugin/marketplace.json and .github/plugin/marketplace.json) must stay in sync with each other.

Deploy Workflow

Follow these steps in order. Stop immediately if any step fails.

Step 1: Pre-flight Checks

Run the preflight script to validate readiness:

node scripts/deploy-preflight.mjs

This checks:

• Current directory is a git repository
• Current branch is master or main
• Working tree is clean (no uncommitted changes)
• /skills directory exists and contains at least one skill
• Detects which surfaces are configured based on existing config files
• Verifies required tools are available for each surface
• Checks that both .claude-plugin/plugin.json and .github/plugin/plugin.json exist; warns if either is missing (auto-created during deploy)
• Reports version consistency across all surface configs, including git tag version

If checks fail, report the problem and suggest a fix. Do not proceed.

Step 2: Analyze Changes

Run the analysis script to understand what changed:

node scripts/deploy-analyze.mjs

This will:

• Inventory all skills in /skills with their names
• Find the last release tag (or handle first release)
• List all commits since that tag
• Count commits by conventional type using anchored regex
• Detect breaking changes (type!: suffix or BREAKING CHANGE in body)
• Show file change statistics
• Print a version bump recommendation

Version bump criteria:

| Bump | When | | --------- | --------------------------------------------------------------- | | Major | Breaking changes — type!: prefix or BREAKING CHANGE body | | Minor | New features — any feat: commits | | Patch | Everything else — fix:, docs:, refactor:, chore:, etc. |

Step 2b: Build Per-Skill Changelog

When the github surface is selected, build a concise, human-readable changelog grouped by skill. This replaces GitHub's auto-generated notes.

Procedure:

1. Identify the previous release tag from deploy-analyze.mjs output (the Last release tag line). If this is the first release, compare against the root commit.
2. For each skill listed in the Skills Changed output, run:

   git diff v{{PREVIOUS}}..HEAD -- skills/{{SKILL_NAME}}/
   

3. Analyze each diff and produce a concise bullet list summarizing user-visible changes per skill. Guidelines:
   • Write each bullet as a short, action-oriented statement (e.g., "Added X", "Fixed Y", "Removed Z").
   • Group by skill as a level-2 heading (## skill-name).
   • Omit internal-only changes (whitespace, line-ending normalization) unless they are the only change.
   • Mention version bumps within skill metadata only if no other substantive changes exist for that skill.
   • Cap at roughly 5–7 bullets per skill; combine minor items if needed.
4. Store the assembled Markdown changelog text for use in Step 9.

Example output format:

## agent-package-manager
- Added subdirectory path and pinned tag dependency examples to template
- Expanded manifest reference with Azure DevOps and GitLab guidance
- Added "APM not installed" troubleshooting section

## agent-skill-deploy
- Made marketplace.json optional for Claude Code surface detection
- Simplified version bump recommendation logic

## github-agentic-workflows
- Added install.md URL reference for agent-assisted setup

If only a single skill changed, omit the heading and use a flat bullet list.

Step 3: Confirm Version and Surfaces with User

Present the analysis summary and ask the user to choose a version and target surfaces.

Use the AskUserQuestion tool for version:

AskUserQuestion:
  question: "Recommended: {{RECOMMENDATION}}. Which version bump for v{{CURRENT}} → v{{NEXT}}?"
  header: "Version"
  options:
    - label: "Major (v{{MAJOR}})"
      description: "Breaking changes — skill removals, renames, config restructuring"
    - label: "Minor (v{{MINOR}})"
      description: "New features — new skills, significant enhancements"
    - label: "Patch (v{{PATCH}})"
      description: "Fixes, docs, refactoring, chore, CI, tests"
    - label: "Cancel"
      description: "Abort the deployment"

Replace {{CURRENT}} with current version, compute {{MAJOR}}, {{MINOR}}, {{PATCH}} by incrementing the relevant segment (reset lower segments to 0).

If user selects Cancel, stop the workflow.

Use the AskUserQuestion tool for surfaces:

AskUserQuestion:
  question: "Detected surfaces: {{DETECTED_SURFACES}}. Which surfaces to deploy to?"
  header: "Surfaces"
  options:
    - label: "All detected ({{DETECTED_SURFACES}})"
      description: "Deploy to every surface that has config files"
    - label: "GitHub release only"
      description: "Create tag and GitHub release"
    - label: "Marketplaces only"
      description: "Update marketplace configs without GitHub release"
    - label: "Custom selection"
      description: "Specify exact surfaces"
    - label: "Cancel"
      description: "Abort the deployment"

Step 4: Dry Run (Recommended)

Before making changes, perform a dry run to verify what will happen:

node scripts/deploy-execute.mjs {{VERSION}} --surfaces {{SURFACES}} --dry-run

Present the dry-run output to the user. The dry run shows:

• Which files will be modified and how
• Which git operations will be performed
• Which surface-specific deploy commands will run
• No files are changed, no git operations are executed

If the dry run reveals issues, address them before proceeding.

Step 5: Bump Versions

Execute the version bump across all detected surface configs:

node scripts/deploy-execute.mjs {{VERSION}} --surfaces {{SURFACES}} --bump-only

IMPORTANT — version sync invariant: The script always bumps every detected config file (plugin.json, marketplace.json, package.json) regardless of which surfaces were selected via --surfaces. This prevents version drift between surfaces. The --surfaces flag only controls which deployment actions run in --push mode.

Auto-creation of missing plugin.json: Before bumping, the script checks that both .claude-plugin/plugin.json (claude-code surface) and .github/plugin/plugin.json (copilot-cli surface) exist. If either file is missing, it is automatically created from the sibling plugin.json (preferred) or from package.json metadata (name, description, version, author, repository, license, keywords) as a fallback. The copilot-cli variant also gets "skills": "skills/".

This updates version fields in:

• .claude-plugin/plugin.json → .version (claude-code surface)
• .claude-plugin/marketplace.json → .plugins[0].version (claude-code surface, only if the file exists)
• package.json → .version (vscode and copilot-cli surfaces)
• .github/plugin/plugin.json → .version (copilot-cli surface, only if the file exists)
• .github/plugin/marketplace.json → .plugins[0].version and .metadata.version (copilot-cli surface, only if the file exists)

If marketplace.json is absent, the plugin is assumed to be listed by a marketplace defined in another repository. Only plugin.json is bumped.

Verify all files were updated correctly by reading them back.

Step 6: Commit Version Bump

Stage and commit all modified config files:

git add -A
git commit -m "Release v{{VERSION}}"

Step 7: Create Git Tag

Tag the release commit:

git tag v{{VERSION}}

Step 8: User Approval Before Push

CRITICAL: Always pause here for user approval.

Present a summary and ask for confirmation.

Use the AskUserQuestion tool:

AskUserQuestion:
  question: "Ready to push v{{VERSION}} and deploy to {{SURFACES}}?"
  header: "Deploy"
  options:
    - label: "Yes — push and deploy"
      description: "Push commits, tags, and run surface-specific deployments"
    - label: "Push only — skip surface deploys"
      description: "Push commits and tags but skip marketplace-specific actions"
    - label: "No — keep local"
      description: "Keep local commit and tag, do not push"

If user selects No, inform them:

• The commit and tag remain local
• To undo: git reset --hard HEAD~1 && git tag -d v{{VERSION}}

Step 9: Push and Deploy

Only after user approves:

1. Save the per-skill changelog from Step 2b to a temporary file:

   echo '{{CHANGELOG}}' > /tmp/release-notes.md
   

2. Run the push and deploy:

   node scripts/deploy-execute.mjs {{VERSION}} --surfaces {{SURFACES}} --push --notes-file /tmp/release-notes.md
   

This performs:

1. git push && git push --tags for all surfaces
2. For github surface: create the release using the per-skill changelog from Step 2b:

   gh release create v{{VERSION}} --title "v{{VERSION}}" --notes-file /tmp/release-notes.md
   

   If --notes-file is not provided, the script falls back to --generate-notes. Always prefer providing the per-skill changelog.

After pushing, print the remote URL and relevant marketplace links.

Error Handling

• No /skills directory: Report that the repository does not follow the expected skill collection layout and stop.
• Missing tools: Report which tools are missing for which surfaces. Suggest installation commands. Allow deployment to surfaces whose tools are available.
• Version mismatch across surfaces: Report the mismatch and suggest running the bump step to synchronize all config files.
• Tag already exists: Report the conflict and suggest either choosing a different version or deleting the existing tag with user confirmation.
• Push failure: Report the error. Do not retry automatically. Suggest checking remote access and authentication.
• Surface deploy failure: If one surface fails, report it but continue with remaining surfaces. Provide per-surface status summary at the end.
• Dry-run divergence: If the actual execution differs from the dry run, pause and report the divergence to the user.