v0.4.0 — Terminal UI, Remote Intent Sources, Lifecycle Hooks & AI Review

Turn AI coding agents into an automated, structured, and repeatable software development workflow.

---

✨ New Features & Workflow Enhancements

• Interactive Terminal User Interface (TUI): Introduced a full-screen, reactive terminal interface built with Ink and React. Features multi-tab navigation across Intents, Tasks, Run Dashboard, Docs, and Config screens (codeforge).
• Onboarding Wizard: Welcomes first-time users with automated CLI detection for installed agents (claude, codex, antigravity/agy, cursor), optional automatic CLI installation, preferred language selection, and an interactive pipeline walkthrough.
• Cross-Platform External Terminal Launcher: Launches the TUI in a dedicated external terminal window by default (Windows Terminal, PowerShell, iTerm2, Kitty, Alacritty, GNOME Terminal) to ensure pristine rendering and avoid collisions with IDE terminals. Supports inline execution via codeforge --inline (-i).
• Remote Intent Sources & Local Materialization (codeforge intent pull): Ingest issues and user stories directly from Linear, GitHub Issues, and ClickUp (or local filesystem) via URL or issue ID. Remote items are materialized locally into .codeforge/intents/<id>.md, keeping the entire remaining development lifecycle 100% offline, local, and reproducible.
• Secure Credential References (apiKeyEnv): Protects secrets by configuring environment variable references (e.g. LINEAR_API_KEY) instead of plain-text tokens in .codeforge/config.yaml.
• Lifecycle Hooks & Auto-Healing Verification Gates: Implemented an extensible hook dispatcher supporting 8 lifecycle events (run.started, run.completed, run.failed, run.deadlock, task.started, task.verify, task.completed, task.failed). Hook classification distinguishes between notify and gate:
  • task.verify gate hooks veto task completion on non-zero exit codes (linters, typecheckers, test suites).
  • Failed hook output is recorded as task diagnostics and automatically injected into the agent's fresh prompt on codeforge task retry for autonomous self-healing.
• Autonomous AI Review Phase: Post-execution review step that evaluates the cumulative Git diff against the intent's requirements and acceptance criteria before marking a run complete. Enforces silence as approval—zero files created indicates approval, while verified defects spawn new TASK-XXX.json files that seamlessly re-queue for execution. Supports --review and --skip-review CLI flags and hotkey v in the TUI.
• Token Efficiency & Context Isolation: Every task executes in a clean, isolated child process, eliminating context bloat and token degradation. System contracts are decoupled from user rules, and prompts use compact JSON schemas with streamed stdin.
• Stage-Specific Customizable Rules: Users can customize project guidelines under .codeforge/rules/ for each distinct stage (planning.md, running.md, review.md, docs.md, docs-update.md). When empty, rules are omitted with zero token overhead.
• Specs → Intents Terminology Migration: Standardized all domain terminology, use cases, CLI commands (codeforge intent create/pull/list/delete), and paths (.codeforge/intents/) from "specifications" to "intents".
• Full CRUD Command Suite:
  • codeforge intent list & codeforge intent delete [name]: Inspect and safely cascade-delete intents, task graphs, executions, and docs.
  • codeforge task delete [intent] [taskId]: Delete individual task JSON files.
  • codeforge docs delete [name]: Delete documentation files and synchronize manifest.json.
• Environment & Configuration Interpolation: Supports root .env and .codeforge/.env files with dynamic ${VARIABLE_NAME} interpolation inside .codeforge/config.yaml.
• Comprehensive Documentation Revamp: Modularized documentation into dedicated guides in docs/ (cli.md, tui.md, intent-sources.md, hooks.md, ai-review.md, token-efficiency.md, configuration.md), updated README.md with explicit "What CodeForge Is / Isn't" positioning, and removed unimplemented deterministic checkers.

---

♻️ Refactor, Architecture & Quality

• Composition Root Dependency Injection: Introduced AppContainer (src/infrastructure/container.ts) providing unified dependency injection across all application use cases and CLI commands.
• Modular TaskScheduler Architecture: Decoupled monolithic scheduler logic into specialized coordinators: TaskExecutor, ReviewCoordinator, SchedulerStateManager, and TaskStorage.
• Process Abstraction (ProcessExecutor): Abstracted external process invocation behind ProcessExecutor and NodeProcessExecutor for consistent cross-platform execution, testing, and real-time streaming.
• TUI State Management: Implemented reactive context providers (ExecutionContext, PlanningContext, ConfigContext, NavigationContext, ContainerContext) for flicker-free terminal updates.
• Real-Time Log Batching & Ticker Singleton: Batched execution log streaming and synchronized animated spinners with SharedSpinnerTicker to minimize terminal redraw overhead.
• Robust Terminal Signal Handling: Added clean SIGINT/SIGTERM handlers and alternate screen buffer restoration across all exit flows.
• Comprehensive Test Suite: 385 automated tests across 40 test files covering intent sources, URL parsing, hook dispatchers, verification gates, scheduler lifecycle, delete use cases, and runners.