claude-workflow-skills

Common workflow skills for Claude Code sessions. Install once via the ali5ter plugin marketplace and invoke with / in any Claude Code session.

Skills

| Skill | Invoke | Description | |---|---|---| | promote | /promote | Full release workflow: commit, push, PR, semver tag, GitHub release, branch cleanup | | audit-plugin | /audit-plugin | Deep review of a Claude Code plugin/skill/agent against official best practices | | audit-standards | /audit-standards | Audit project against personal dev standards in ~/.claude/CLAUDE.md | | improve | /improve | Analyse project for bugs, feature gaps, docs, security, competitive opportunities, and monetisation — file as GitHub issues | | triage | /triage | Validate all open issues, close invalid ones, flag complex ones for planning, fix and promote the rest | | review-pr | /review-pr | Review an open PR against coding standards, security, and test coverage, then post a structured review and approve or request changes |

Installation

Add the marketplace (once)

/plugin marketplace add ali5ter/claude-plugins

Install this plugin

/plugin install claude-workflow-skills@ali5ter

Update

/plugin update claude-workflow-skills@ali5ter

Usage

Each skill is invoked with its / prefix from any Claude Code session:

/promote
/audit-plugin
/audit-standards
/improve

All skills have disable-model-invocation: true — they must be invoked explicitly via their / prefix. Claude will not fire them automatically, even if the conversation matches a trigger phrase. This prevents accidental side effects (commits, issue creation, PR reviews) without explicit user intent.

Recognised trigger phrases (for reference — always invoke manually):

| Skill | Trigger phrases | |---|---| | /promote | "ship this", "release", "get back to main", "tag and release" | | /audit-plugin | "audit this plugin", "review this skill", "check this agent" | | /audit-standards | "audit standards", "check standards compliance", "audit against settings" | | /improve | "improve this", "analyse this project", "find improvements", "fill the backlog" | | /triage | "triage issues", "validate issues", "fix issues", "work through the backlog" | | /review-pr | "review this PR", "check the PR", "look at PR #N", "review pull request" |

Skill details

/promote

Runs the full release workflow for the current project:

1. Commits any uncommitted changes with a conventional commit message
2. Pushes to remote
3. Creates and merges a PR if on a feature branch (with auto-merge)
4. Determines the next semver version from conventional commits since the last tag
5. Creates an annotated git tag and GitHub release with generated release notes
6. Cleans up merged feature branches
7. Returns to a clean main

Requires gh (GitHub CLI) to be authenticated.

/audit-plugin

Performs a deep review of the Claude Code addon defined in the current project:

• Reads plugin.json, SKILL.md, and agent .md files
• Fetches current best-practice documentation
• Checks frontmatter fields, tool declarations, description format, and body quality
• Creates GitHub issues for each finding
• Writes a prioritised fix plan to the project CLAUDE.md

/audit-standards

Audits the current project against the seven development principles in ~/.claude/CLAUDE.md:

• Codify, Don't Document
• Bash Script UX with pfb
• Markdown Standards
• Professional Documentation Tone
• Version Control Everything

Creates GitHub issues for non-compliance and writes a prioritised fix plan to CLAUDE.md.

/improve

Analyses the current project from multiple angles and files GitHub issues for every finding:

• Code quality and bugs — logic errors, anti-patterns, fragile assumptions
• Feature completeness — gaps relative to the project's stated purpose
• Documentation — missing, outdated, or misleading content
• Security — injection risks, exposed secrets, over-permissioned scopes
• Competitive landscape — what similar tools offer that this project doesn't
• Monetisation — sponsorship, premium tiers, marketplace listings, companion products

Produces a prioritised summary table with top issues and highest-leverage opportunities.

Note: Steps 6 and 7 (competitive landscape and monetisation) require WebSearch and WebFetch permissions. In restricted permission mode, approve these when prompted or those steps will be skipped.

/triage

Works through every open GitHub issue and resolves them:

1. Fetches and classifies all open issues as invalid, complex, or actionable
2. Closes invalid issues with a documented comment explaining why
3. Surfaces complex issues for planning discussion before proceeding
4. Fixes each actionable issue with a minimal, scoped change
5. Runs /promote to commit, tag, and release

In non-interactive mode, stops after surfacing complex issues rather than waiting for input.

/review-pr

Reviews an open pull request and posts a structured GitHub review:

1. Fetches the PR diff and metadata via gh pr diff
2. Evaluates correctness, coding standards, security, test coverage, documentation, and scope
3. Drafts a structured review (must-fix findings, suggestions, positive notes, and verdict)
4. Posts the review via gh pr review --approve or gh pr review --request-changes

Defaults to the current branch's open PR; pass a PR number to review any PR.

Architecture

Skills are defined in skills/<name>/SKILL.md with YAML frontmatter. The promote skill uses shell command injection (!`command`) to inject live git state before Claude processes the instructions, giving it accurate context about the current branch, last tag, and recent commits.

Support

If these skills save you time, consider sponsoring on GitHub.

License

MIT © Alister Lewis-Bowen