Juni Skills Suite - A unified Claude Code plugin combining feature development workflows (/juni:cook) and context safety tools (/juni:guard).
| Command | Description |
|---|---|
/juni:cook |
Structured feature development with planning phases |
/juni:cook-menu |
Interactive artifact management |
/juni:cook-stats |
Analytics and insights for cook artifacts |
/juni:sous-chef |
Background monitoring for cooking discipline |
/juni:inspect |
Post-implementation sanitation inspection |
/juni:guard |
Epistemic safety for JSON data in prompts |
# Add the marketplace (run /plugin, select "Add Marketplace", enter: PJuniszewski/juni-skills-marketplace)
# Install and enable
claude /plugin install juni-skills:juni
claude /plugin enable juninpx skills add PJuniszewski/claude-cook# Check available commands
/help
# Test cook
/juni:cook Add a simple feature --dry-run
# Test guard
/juni:guard '[{"test": 1}]'/juni:cook prevents shipping raw code by requiring structured planning, review phases, and documented decisions before implementation begins.
Full governance cooking with all review phases:
/juni:cook Add SSE streaming blocks --well-done
/juni:cook Add user authentication with OAuth
/juni:cook Implement real-time notifications
Speed-optimized for low-risk changes:
/juni:cook Fix crash in SettingsActivity --microwave
/juni:cook Fix null pointer in payment handler --microwave
/juni:cook Update error message text --microwave
Each /juni:cook run produces an artifact in cook/<slug>.<date>.cook.md:
# Cooking Result
## Dish
Add SSE streaming for real-time updates
## Status
well-done
## Cooking Mode
well-done
## Ownership
- Decision Owner: @engineer
- Reviewers: auto
- Approved by: Product on 2026-01-07
## Product Decision
Approved
- Reason: Aligns with Q1 real-time features roadmap
## Pre-mortem
1. Connection drops silently -> mitigation: heartbeat + reconnect logic
2. Memory leak from unclosed streams -> mitigation: cleanup on unmount
3. Server overload from many connections -> mitigation: connection pooling
## Patch Plan
- Files to modify:
1. src/api/streaming.ts - new SSE client
2. src/hooks/useStream.ts - React hook wrapper
- Tests to run: streaming.test.ts, integration/sse.test.tsThese topics automatically escalate to --well-done:
| Topic | Reason |
|---|---|
| auth / permissions / crypto | Security-critical |
| schema / migrations / storage | Data integrity risk |
| public API contracts | Breaking change risk |
| UI flow changes | UX impact |
| payments / purchase / paywall | Financial/compliance risk |
| Status | Meaning |
|---|---|
raw |
Requested, not evaluated |
cooking |
In progress |
blocked |
Blocker identified |
needs-more-cooking |
Rejected or incomplete |
well-done |
Approved, ready to implement |
ready-for-merge |
Post QA/Security |
plated |
Shipped |
/juni:guard prevents LLMs from reasoning with unjustified certainty when input data is incomplete.
- Lossless reduction - Minify, columnar transform, remove nulls
- Token counting - API or heuristic fallback
- Decision engine - ALLOW / SAMPLE / BLOCK
- Intelligent trimming - First + last + evenly-spaced sampling
- Forensic detection - Warns when specific record queries detected
# Analyze a file
/juni:guard my_data.json
# Analyze inline JSON data
/juni:guard '[{"id": 1}, {"id": 2}]'
# Force through despite warnings
/juni:guard my_data.json --force
# Check if forensic query is safe
/juni:guard logs.json --mode forensics
# Use larger token budget
/juni:guard data.json --budget-tokens 5000| Mode | Sampling | Use Case |
|---|---|---|
analysis |
Allowed | "What categories exist?", "Price range?" |
summary |
Aggressive | "Describe the data structure" |
forensics |
BLOCKED | "Why did request id=X fail?" |
| Variable | Default | Description |
|---|---|---|
TOKEN_GUARD_MIN_CHARS |
6000 |
Below = always allow |
TOKEN_GUARD_WARN_CHARS |
15000 |
Above = warn |
TOKEN_GUARD_HARD_LIMIT_CHARS |
100000 |
Above = hard block |
TOKEN_GUARD_MODEL |
claude-sonnet-4-20250514 |
Model for token counting |
TOKEN_GUARD_PROMPT_LIMIT |
3500 |
Default token budget |
Interactive artifact management:
- Compare artifacts
- Validate against requirements
- View status dashboard
/juni:cook-stats # Show overall statistics
/juni:cook-stats --since 2026-01-01 # Filter by date
/juni:cook-stats search "authentication" # Search artifacts
/juni:cook-stats similar src/auth.ts # Find similar by files
Background monitoring:
/juni:sous-chef monitor # Detect uncooked sensitive changes
/juni:sous-chef drift cook/feature.cook.md # Compare plan vs implementation
/juni:sous-chef postmortem cook/feature.cook.md # Analyze predictions
/juni:sous-chef suggest # Get governance suggestions
Post-implementation sanitation inspection - code review that verifies actual git diffs match the cook artifact plan:
/juni:inspect # Inspect most recent well-done artifact
/juni:inspect cook/feature.cook.md # Inspect specific artifact
/juni:inspect artifact --commit abc123 # Inspect specific commit against artifact
/juni:inspect --commit abc123 # Pure code review (no artifact)
/juni:inspect --surprise # Force surprise inspection mode
Three Modes:
| Mode | Description |
|---|---|
| Artifact + Auto-detect | Find artifact, find related commits, compare plan vs actual |
| Artifact + Manual commit | Use specified artifact and commit |
| Commit only | Pure code review (hygiene + safety, no recipe compliance) |
What it checks:
- Recipe Compliance - Plan vs actual implementation (planned files, decisions, non-goals)
- Hygiene - Error handling, edge cases, code consistency in the diff
- Safety - Input validation, injection vectors, auth checks in the diff
High Signal Only: Inspired by Anthropic's code-review, inspect only flags issues with high confidence:
- Syntax errors, missing imports
- Clear logic errors
- Deviation from artifact plan
- Obvious security vulnerabilities
Surprise Mode (--surprise):
Stricter inspections for high-risk situations:
| Aspect | Regular | Surprise |
|---|---|---|
| Confidence threshold | >= 75% | >= 60% |
| Agent model | Sonnet | Opus |
| Additional checks | 3 agents | + CLAUDE.md compliance |
Use for: auth changes, payments, schema migrations, large diffs, pre-release audits.
Edge Case Handling:
- Multiple artifacts? Interactive selection
- No commits found? Option for manual SHA or pure code review
- Files don't match plan? Confirmation before proceeding
The inspection report is appended to the cook artifact with violations and recommendations.
Create project-specific chefs in .claude/agents/:
.claude/agents/
product_chef.md # Scope decisions
ux_chef.md # UX review
qa_chef.md # Test planning
security_chef.md # Security audit
Project chefs override system-wide defaults.
/juni:cook extracts project context during Phase 0. The more context available, the better the output.
Recommended project setup:
| File | Purpose |
|---|---|
CLAUDE.md |
Project rules, constraints, conventions |
README.md |
Project overview, architecture summary |
docs/ |
Architecture decisions, API specs, ADRs |
.claude/agents/ |
Project-specific review chefs |
| Document | Description |
|---|---|
| CLAUDE.md | Project rules for this repository |
| CHEF_MATRIX.md | Who does what - chef responsibilities |
| COOK_CONTRACT.md | Required sections for valid artifacts |
| ANTI_PATTERNS.md | What /juni:cook is NOT |
| COMPARISON.md | When to use what |
Compare two cook artifacts to see what changed:
# Compare two artifacts
./scripts/cook-diff cook/feature-a.cook.md cook/feature-b.cook.md
# Show changelog entries since a date
./scripts/cook-diff cook/feature.cook.md --since 2026-01-01Output shows:
- Added sections (new in second file)
- Removed sections (missing in second file)
- Modified sections (content changed)
- Summary of total changes
Each artifact includes a ## Changelog section for tracking version history within the file.
If you had the old plugins installed:
# Remove old plugins
claude /plugin uninstall juni-skills:cook
claude /plugin uninstall juni-skills:context-guard
# Install unified plugin
claude /plugin install juni-skills:juni
claude /plugin enable juniCommand changes:
| Old | New |
|---|---|
/cook |
/juni:cook |
/cook-menu |
/juni:cook-menu |
/cook-stats |
/juni:cook-stats |
/sous-chef |
/juni:sous-chef |
/guard |
/juni:guard |
Validate cook artifacts against mode-specific requirements:
# Validate an artifact
./scripts/cook-validate cook/feature.cook.md
# Verbose output (show all checks)
./scripts/cook-validate cook/feature.cook.md --verbose
# JSON output (for CI)
./scripts/cook-validate cook/feature.cook.md --jsonChecks include:
- Required sections present (scope, pre-mortem, tests, etc.)
- No TBD/TODO placeholders
- Ownership assigned
- Minimum test cases defined
- Rollback plan documented
Use /cook-menu for interactive artifact management (validate, compare, view status).
MIT License. See LICENSE.