WTF-P v0.5.0 — Multi-Runtime Parity
36 commands and 11 specialized agents now run on Claude Code, Gemini CLI, and OpenCode.
Install
npx wtf-p --global # Claude Code (default)
npx wtf-p --global --gemini # Gemini CLI
npx wtf-p --global --opencode # OpenCodeHighlights
Multi-Runtime Support
- Gemini CLI — 36 TOML-format commands + 11 agents
- OpenCode — 36 Markdown commands + 11 agents
- MANIFEST-based installer with per-runtime file tracking and clean uninstall
GSD-Inspired Agent Architecture
- 7 specialized agents: section-planner, section-writer, section-reviewer, plan-checker, research-synthesizer, prose-polisher, argument-verifier
- Thin orchestrator pattern: 11 commands refactored from monoliths to agent-spawning orchestrators (54% code reduction)
- Structured returns: COMPLETE / CHECKPOINT / BLOCKED for deterministic routing
Quality Loops
- Plan-Check-Revise: Pre-write validation across 7 dimensions with up to 3 revision iterations
- Goal-Backward Verification: Post-write checks that claims match plans and evidence
Model Profiles
quality/balanced/budgetprofiles route 11 agents to appropriate models (opus/sonnet/haiku)- Configurable via
config.json— no code changes needed
8 New Commands
verify-work, execute-outline, settings, add-todo, check-todos, update, audit-milestone, plan-milestone-gaps
Wave-Based Parallel Writing
Section plans include wave and depends_on for parallel execution of independent sections.
New Features
- Context primer — Section-specific context extraction for journal-scale papers
- Preference inheritance — Global style/citation defaults with per-project overrides (
base-prefs.yaml) - Checkpoint system — Git-tagged save/restore/list for paper state
/wtfp:quick— Minimal-ceremony writing for quick tasks- Config extensions —
model_profile,workflow.research,workflow.plan_check,workflow.verifier,parallelization.enabled
Fixed
preflight.jstemplate path resolutioncreate-outline.wcn.mdstep name mismatch- Dual installation (global + local) conflict detection
- 5 install/uninstall edge cases found via E2E testing
- Stale
research-gap.wcn.mdsynced with verbose counterpart
Stats
- 788 tests passing across 7 suites (sanity, paths, linter, WCN integrity, dry-run, features, installer)
- 249 files in npm tarball, 302 kB packed
- 36 commands × 3 runtimes = 108 command files
- 11 agents × 3 runtimes = 33 agent files
Full changelog: https://github.com/akougkas/wtf-p/blob/main/CHANGELOG.md
Built at the Gnosis Research Center at Illinois Tech.