v1.12.0 — address-pr-review skill + tracked fix loop + spec alignment

Highlights

• New skill: address-pr-review bundled into the code-review plugin — validates GitHub PR comments across all three surfaces (reviewThreads, issue-level comments, reviews.body), applies fixes for valid ones, replies with reasoning, and resolves threads. Push happens before reply so cited SHAs are always live. No force pushes, no --no-verify.
• code-review overhaul: tracked fix loop (TaskCreate/TaskUpdate with post-fix verification), Failure & Resumption Protocol (Retry / Skip / Abort), Next-Pass Gate after each iteration, dependency grounding against pinned manifest versions (Node, Python, Go, Java, Kotlin, Rust, Ruby, PHP, Elixir + runtime pinning files), base-branch auto-detection, Dependency Compatibility added as 10th gap category. Body split into 4 references files. AskUserQuestion invocations mandated verbatim at every gating step.
• skill-creator brought in line with the current Claude Code skill spec: SKILL.md, quick_validate.py, and init_skill.py template now use description + when_to_use, accept the 15 official frontmatter fields, enforce the 1,536-char combined cap. Fixes an invalid-YAML bug in the generated skill template.
• All other plugins (ultrathink, database, ideate, interview): frontmatter migrated to description + when_to_use; non-spec fields (license, invocation: user) removed.
• CI: new skill-validate job runs quick_validate.py against every SKILL.md (with exit-code propagation fixed).

Per-plugin versions

Plugin	Old → New
code-review	1.0.1 → 1.1.0 (minor — new skill)
skill-creator	1.0.1 → 1.1.0 (minor — body + validator + init template)
ultrathink	2.0.2 → 2.0.3 (patch)
database	1.2.2 → 1.2.3 (patch)
ideate	1.0.0 → 1.0.1 (patch)
interview	1.0.0 → 1.0.1 (patch)

Installation

```bash
claude plugin marketplace add husniadil/ekstend
claude plugin install code-review@ekstend
```

What's Changed

• PR #2 — Release v1.12.0 — #2

🤖 Generated with Claude Code

v1.11.0

New Plugin: Interview

Deep requirements gathering skill that interviews users through thoughtful, in-depth questions before implementation.

Features

• CDEEPER Workflow - 7-phase interview process:

  • Contextualize: Explore codebase before asking questions
  • Discover: Understand core intent
  • Explore: Technical implementation questions
  • Edge: Edge cases and boundaries
  • Prioritize: Trade-offs and constraints
  • Experience: UX/DX considerations
  • Ready: Summary and plan mode transition

• Skip-to-Summary Flow - Jump to summary at any point with configurable styles:

  • Minimal: List what was discussed
  • Inferred: Make reasonable assumptions
  • Critical first: Ask 2-3 critical questions before summarizing

• Question Tracking & Deduplication - Avoid re-asking covered topics

• Confidence Scoring - Percentage-based confidence on each requirement (90-100% explicit, 70-89% uncertain, etc.)

• Hybrid Question Ordering - Domain-based start with adaptive reordering based on answers

• Multi-session Persistence - Resume interviews with full phase tracking

Installation

claude plugin install interview@ekstend

Usage

/interview
"Interview me about this feature"
"Let's deep dive into requirements before implementing"

---

🤖 Generated with Claude Code

v1.10.2

What's Changed

Ultrathink Plugin v2.0.2

Auto-correct shell escaped quotes in JSON arguments

Fixed a common issue where LLMs would pass incorrectly escaped JSON arguments to the ultrathink CLI, causing validation errors.

Features

• Auto-correct shell escaping mistakes like '[{\"id\":\"A1\"}]' to valid JSON
• Handle nested escaping where text contains escaped quotes (e.g., '[{\"text\":\"has \\\"quotes\\\"\"}]')
• Refactored _parse_json_list() for cleaner code and better maintainability

Technical Details

• Added _fix_shell_escaped_quotes() function to detect and fix common shell escaping patterns
• Added _ParseResult class and _try_parse_as_list() helper for cleaner parsing logic
• All 134 tests passing

---

🤖 Generated with Claude Code

v1.10.1

Bug Fixes and Improvements

code-review plugin (v1.0.1)

• Added gap validation step - Now validates agent-reported gaps by reading actual code before presenting to user
• Added anti-pattern warning: "Trusting agent claims blindly"
• Updated description to include holistic codebase review mode

skill-creator plugin (v1.0.1)

• Fixed script paths to use ${CLAUDE_PLUGIN_ROOT}/skills/skill-creator/scripts/
• Fixed package_skill.py path comments
• Added documentation for allowed YAML frontmatter properties (allowed-tools, metadata)

ultrathink plugin

• Fixed ultrathink.py version mismatch (synced to 2.0.1)
• Updated AGENTS.md line counts (~980 lines, 134 tests)

---

🤖 Generated with Claude Code

v1.10.0

New Plugin: code-review

An iterative, human-in-loop code review skill that identifies gaps, presents them for user selection, and fixes them in cycles until clean.

Features

• Two Review Modes: PR diff review and holistic codebase audit
• Business Logic Awareness: Understands domain rules, not just code correctness
• End-to-End Integration Tracing: Traces frontend → backend → database → external APIs
• 9 Gap Categories: Business Logic, Integration, Logic, Security, Performance, Error Handling, Style, Tests, Docs
• Human-in-Loop: Uses AskUserQuestion for gap selection
• Iterative Loop: Review → Fix → Re-review until clean

Installation

claude plugin install code-review@ekstend

Usage

review my PR
review this codebase
audit the project for gaps

Other Changes

• Fixed duplicate ideate section in README.md

---

🤖 Generated with Claude Code

v1.9.3

ultrathink v2.0.1

Bug Fixes

• Support Python literal syntax for JSON arguments - Fixes shell argument mangling where double quotes get converted to single quotes when passing --assumptions, --depends-on, or --invalidates via CLI. Added ast.literal_eval fallback parsing to handle Python-style syntax like [{'id':'A1',...}].

Full Changelog

v1.9.2...v1.9.3

v1.9.2 - English-only Keywords

What's Changed

Database Plugin v1.2.2

• Removed Indonesian phrases from skill description
• English-only trigger keywords: "check data", "get data", "fetch data", "look up", "retrieve", "from database", "in database"

Version Updates

• Marketplace: 1.9.1 → 1.9.2
• Database plugin: 1.2.1 → 1.2.2

v1.9.1 - Aggressive Skill Trigger

What's Changed

Database Plugin v1.2.1

• Aggressive skill trigger - Added "ALWAYS USE THIS SKILL" prefix
• More keywords - orders, users, cek data, lihat data, di db
• Indonesian phrases - cek order, lihat user, ambil data, di database

This should improve auto-detection when users ask database-related questions without explicitly invoking the skill.

Version Updates

• Marketplace: 1.9.0 → 1.9.1
• Database plugin: 1.2.0 → 1.2.1

v1.9.0 - Query Workflow & Better Triggers

What's New

Database Plugin v1.2.0

• Expanded skill triggers - Now triggers on "db", "check", "look up", "records", "data lookup"
• Critical Rules section - Forbids guessing table/column names without schema exploration
• Query Workflow section - Mandatory 4-step process for unfamiliar databases:
  1. List tables (SHOW TABLES, \dt, .tables)
  2. Identify target table from actual list
  3. Describe table structure before querying
  4. Build query using actual column names

Why This Matters

Prevents incorrect queries caused by assuming ORM naming conventions (e.g., Order vs orders, table_name vs service_type).

Version Updates

• Marketplace: 1.8.0 → 1.9.0
• Database plugin: 1.1.0 → 1.2.0

v1.8.0 - Redis Support

What's New

Database Plugin v1.1.0

• Redis support - Added Redis as fourth database type
• New references/redis.md with complete CLI documentation
• Support for redis://, rediss:// URLs and REDIS_* environment variables
• All Redis data structures: strings, hashes, lists, sets, sorted sets, streams
• Redis-specific safety rules (FLUSHDB, FLUSHALL, DEL patterns)
• Key exploration with SCAN (production-safe)
• TLS/SSL connection support
• Link to official Redis docs: https://redis.io/docs/latest/commands/

Version Updates

• Marketplace: 1.7.0 → 1.8.0
• Database plugin: 1.0.0 → 1.1.0