claude-skills
Claude Code plugins by Ava Barron.
Install
Add this marketplace once, then install any plugin:
# In Claude Code
/plugin marketplace add asbarron/claude-skills
# Install a plugin
/plugin install <plugin-name>@asbarron-skills
CLI equivalent:
claude plugin marketplace add --scope user asbarron/claude-skills
claude plugin install <plugin-name>@asbarron-skills
Plugins
pr-review
Calibrated PR reviews via Claude Code. Run /pr-review in any repo with an open PR — Claude reads the diff, calibrates to the repo's current state, and posts inline comments + an overall verdict to GitHub under your own account via gh CLI.
Use --deep for a multi-lens review: four parallel agents (code quality, security, architecture, git history) with research-backed confidence scoring that filters false positives.
Prerequisites: gh CLI installed and authenticated (gh auth login).
Usage
/pr-review # current branch's PR
/pr-review 42 # specific PR (bare, #42, or full URL)
/pr-review 42 --dry-run # preview without posting
/pr-review 42 --approve # post as approval (if no blocking issues)
/pr-review 42 --deep # multi-lens review with confidence scoring
/pr-review 42 --deep --dry-run
How it works
Normal mode (default):
- Reads repo context (age, structure, test/CI presence) to calibrate review depth
- Fetches the diff and existing comments
- Applies the reviewer persona — Conventional Comments taxonomy, max 2 nitpicks, no style comments if a linter is configured
- Posts inline comments + overall verdict via
gh api under your GitHub identity
Deep mode (--deep):
- Same context gathering, plus git history for touched files
- Launches four review lenses in parallel:
- General — correctness, edge cases, test coverage, efficiency
- Security — exploitable vulnerabilities with concrete attack scenarios
- Architecture — layer violations, pattern consistency, module boundaries
- Git History — reverted fixes, removed guards, churn patterns
- Runs a confidence scoring pass that deduplicates, verifies assumptions, and filters false positives
- Posts only Medium+ severity findings (Critical, High, Medium)
Labels
| Label | Meaning |
|---|
issue: | Must fix — correctness, security, data loss |
suggestion: | Take it or leave it |
question: | Genuine confusion, not a criticism |
nitpick: | Minor style/naming, your call |
praise: | Something done well |
Append [non-blocking] to an issue: that matters but isn't merge-blocking.
pro-plan
Plan and TDD review via Claude Code. Run /pro-plan on any implementation plan — Claude detects the tech stack, web-searches for canonical resources (books, official docs), reads CONTRIBUTING.md as a source of truth, and runs four parallel review lenses. Then it asks you targeted questions and produces a revised plan.
Prerequisites: None.
Usage
/pro-plan # most recent Claude Code plan
/pro-plan path/to/plan.md # specific plan or TDD
/pro-plan path/to/plan.md --review-only # review + questions, no revision
/pro-plan path/to/plan.md --revise-only # restructure/improve, no review
Revised plans are written to {original-name}-pro.md in the same directory.
How it works
Phase 1 — Review:
- Reads the plan and classifies its maturity (sketch, standard, RFC/TDD)
- Detects the tech stack from plan text and repo signals (
go.mod, package.json, etc.)
- Web-searches for canonical resources per technology (e.g., "The Go Programming Language", DDIA, official docs)
- Reads CONTRIBUTING.md (if it exists) as an additional source of truth
- Launches four review lenses in parallel:
- DRY — duplicated logic, repeated steps, reinventing existing code
- Best Practices — anti-patterns and deprecated approaches per tech stack
- Structure — phase ordering, dependencies, parallelization opportunities
- Completeness — missing error handling, testing, rollback, security
- Runs a review synthesizer that deduplicates, scores, and filters findings
- Asks you up to 5 targeted questions about gaps and ambiguities
Phase 2 — Revise:
- Incorporates review findings and your answers into a revised plan
- Preserves your original voice and structure — only changes what needs changing
- Marks modifications with
<!-- pro-plan --> comments so you can see what changed
Labels
