Claude Code plugins

Two Claude Code plugins bundling slash-command skills: audit (review workflows) and workflow (project workflow automation). Each skill is invoked explicitly via its slash command. See each plugin's README for skill-level details. See the Claude Skills documentation for how Claude Skills work.

Install

Add the marketplace:

claude plugin marketplace add hebstr/claude-code-plugins

Then install plugins individually:

claude plugin install audit@hebstr
claude plugin install workflow@hebstr

Plugins

audit

Review workflows for code, Claude Code skills and MCP servers. Covers point-by-point walkthroughs, full-project sweeps, and adversarial cross-model review with circularity detection.

Skill	Purpose
walkthrough	Interactive, point-by-point walkthrough of any audit report. Re-evaluates findings, proposes fixes, and checks impacted files for regressions
sweep	Full project review. Spawns parallel agents (architecture, quality, tests, docs) and consolidates findings into one deduplicated report
blindspot	Reviewer orchestrator that detects circularity (reviewer shares its target's codebase or model family) and injects cross-model judging via OpenRouter
skill-adversary	Adversarial reviewer for Claude Code skills. Surfaces trigger edge cases, instruction ambiguities, cross-file coherence issues, and gaps
mcp-adversary	Adversarial reviewer for MCP servers. Surfaces tool-selection ambiguity, schema anti-patterns, semantic drift, and undocumented workflow dependencies

See audit/README.md for per-skill details.

workflow

Project workflow automation: file consistency sweeps, cross-repo synchronization, prose editing, session handoffs, source-backed recommendations, and documentation layout audits.

Skill	Purpose
sync	Scans all project files, identifies those stale relative to recent changes, and updates them. Always runs a cross-repo semantic consistency pass with parallel agents
write	Strips AI writing patterns and rewrites prose to read naturally. Routes to a French or English reference per text language. Includes a bilingual review mode (FR-EN parity, typo, faux amis)
continue	Flushes durable facts to memory, updates .claude/PLAN.md, and prints a continuation prompt. No handoff document: PLAN.md and memory are authoritative
reco	Deep recommendation backed by external sources. Parallel agents pull official docs (WebFetch) and community practice (WebSearch); the skill synthesizes a structured recommendation with verified citations
doc-structure	Audits documentation layout (CLAUDE.md vs README.md), proposes verbatim migrations of misplaced prose, and updates the CLAUDE.md index

See workflow/README.md for per-skill details.

Requirements

Required for both plugins:

• Claude Code with plugin marketplaces enabled
• git (review targets and workflow:sync operate on git state)

Optional, per feature. Skills degrade gracefully: a missing dependency disables the affected feature and is reported. The rest of the skill runs.

Dependency	Used by	Unlocks
OPENROUTER_API_KEY env var (+ jq CLI)	audit:blindspot, audit:walkthrough	Cross-model judging and cross-provider validation via OpenRouter
Ouroboros plugin	audit:walkthrough	Consensus QA verdicts, evaluate + drift checks, lateral-think rescue
critical-code-reviewer (posit-dev/skills)	audit:sweep, audit:walkthrough	Specialist code-review agent. Default reviewer in audit:sweep; recommended --reviewer for audit:walkthrough on code targets. Falls back to inline review when absent
r-lib skills testing-r-packages, r-package-development, cran-extrachecks (posit-dev/skills)	audit:sweep on R projects	R-specific test, package, and CRAN-readiness audits
gh CLI	audit:walkthrough	PR body as evaluate context (falls back to latest commit message)

License

MIT