v0.20.2 - Fix CLI log path mismatch

Fixed

• CLI log path mismatch (GitHub Issue #6): Displayed log paths now match actual files created
  • Single execution_timestamp generated once and passed through entire execution chain
  • Loop state stores timestamp for resume consistency
  • Both foreground/background loop modes show correct loop_log path
  • Claude subagent mode now creates logs in ~/.claude/cli-logs/ (was /tmp/)

Added

• --execution-timestamp internal CLI argument for timestamp consistency across loop iterations
• Loop log metadata header (prompt, model, timestamp, max iterations, CWD, worktree, branch)
• 6 new unit tests for log path consistency across all execution modes

Changed

• run_verification_loop() and run_verification_loop_background() now accept execution_timestamp param
• Loop state schema extended with execution_timestamp field for resume support

---

Full Changelog: v0.20.1...v0.20.2

v0.20.0 - OpenCode CLI Support

What's New

OpenCode CLI Support for Z.AI GLM-4.7

This release adds support for OpenCode CLI as the recommended way to run Z.AI GLM-4.7 models.

Why OpenCode?

• Native Z.AI support with proper GLM-4.7 message format handling
• Fixes GitHub Issue #4 (Z.AI GLM-4.7 compatibility with Codex CLI)
• The existing zai model (via Codex profile) is retained but may have issues with certain prompts

Usage

# Recommended: Use OpenCode for Z.AI
/run-prompt 005 --model opencode

# Alternative: Still available via Codex
/run-prompt 005 --model zai

Changes

• Added: opencode model for Z.AI GLM-4.7 via OpenCode CLI
• Added: PTY wrapper support for CLIs that require pseudo-terminals
• Added: needs_pty flag in model configuration
• Added: 7 new unit tests for model configuration and PTY command wrapping
• Fixed: Z.AI GLM-4.7 compatibility issues (GitHub Issue #4)
• Updated: Model reference tables with opencode as recommended Z.AI option

Update

claude plugin uninstall daplug@cruzanstx
claude plugin marketplace update cruzanstx
claude plugin install daplug@cruzanstx

Full changelog: CHANGELOG.md

v0.18.0 - Sprint Planning

🚀 New Feature: /sprint Command

Automated sprint planning from technical specifications!

What's New

The /sprint command transforms a technical spec into an executable sprint plan:

# From a spec file
/sprint ./docs/technical-spec.md --worktree --loop

# From inline text
/sprint "Build a REST API with auth, CRUD, and admin dashboard"

# Dry run to preview
/sprint ./spec.md --dry-run

Features

• Spec Analysis - Identifies components, dependencies, and relationships
• Prompt Generation - Creates numbered prompts with proper structure
• Dependency Graph - Builds and visualizes component dependencies
• Model Assignment - Assigns optimal AI models based on task type
• Execution Plan - Groups prompts into phases with parallelization
• State Management - Persistent .sprint-state.json for long-running sprints

Sub-commands

/sprint status    # View current sprint status
/sprint add       # Add a prompt to sprint
/sprint remove    # Remove a prompt
/sprint replan    # Re-analyze dependencies
/sprint pause     # Pause sprint
/sprint resume    # Resume from last phase
/sprint cancel    # Cancel and cleanup
/sprint history   # Show sprint history

Files Added

• commands/sprint.md - Command definition
• skills/sprint/SKILL.md - Skill definition
• skills/sprint/scripts/sprint.py - Core implementation (1,428 lines)
• skills/sprint/tests/test_sprint.py - Test suite

---

Closes #3

🤖 Generated with Claude Code

v0.17.1

Bug Fixes

• fix(check-updates): Use curl to fetch actual remote version from GitHub
  • Previous approach used git archive --remote which doesn't work with GitHub
  • Now fetches plugin.json directly from raw.githubusercontent.com

v0.17.0

Bug Fixes

• fix(executor): Improve verification marker reliability for Gemini models (#2)
  • Moved completion marker instructions outside XML blocks into a dedicated MANDATORY section
  • Added DAPLUG_INSTRUCTIONS_END sentinel to prevent false-positive detection from echoed prompts
  • Backward compatible with existing loop states

Testing

• Added unit tests for marker detection edge cases (test_executor_markers.py)
• Verified with live Gemini execution - completion marker now detected on first iteration

Contributors

• Codex (gpt-5.2-codex)
• Claude (claude-opus-4-5)