Release v2.2.0 (#71)

* Add Claude Code Router + Codex CLI features, extract dashboard, restructure docs

Container:
- Add claude-code-router feature for LLM provider routing (DeepSeek, Gemini, OpenRouter)
- Add codex-cli feature for OpenAI Codex terminal agent
- Add defaults/ directory with factory config templates
- Update ccusage feature with Codex support
- Update auth scripts for new provider API keys

Dashboard:
- Extract dashboard/ to separate CodeDirective repository
- Add dashboard/ to .gitignore (deprecated in monorepo)

Documentation:
- Restructure docs: getting-started → start-here, features/customization → use/customize
- Move plugins/ under extend/ section
- Add new reference pages: agents, skills, cli-tools, environment-variables
- Consolidate architecture and troubleshooting in reference/

Housekeeping:
- Gitignore temp directories and screenshots

* Remove dashboard feature from devcontainer

Dashboard extracted to separate CodeDirective repository. Remove:
- codeforge-dashboard devcontainer feature
- Dashboard port forwarding and config from devcontainer.json
- Dashboard documentation page and sidebar references
- Dashboard mentions from README, CLAUDE.md, and related docs

* Remove deprecated container/.codeforge directory

Configuration migrated to container/.devcontainer/defaults/codeforge/.
The .codeforge directory at package root was a development artifact that
should not ship with the npm package.

* Add behavioral rules for Claude Code sessions

New rules in defaults/codeforge/config/rules/:
- auto-memory.md: memory constraints and staleness cleanup
- zero-tolerance-bugs.md: bugs always in scope, must be fixed
- scope-discipline.md: only user defines scope
- explicit-start.md: never start without clear instruction
- plan-presentation.md: compressed overview before full plan
- surface-decisions.md: surface assumptions before acting

* Update hook gate path to ~/.claude/disabled-hooks.json

All plugin scripts now check ~/.claude/disabled-hooks.json instead of
.codeforge/config/disabled-hooks.json for the hook disable list.

* Update defaults configuration

- settings.json: update to opus-4-7, reduce thinking tokens (31999),
  disable adaptive thinking, add effort_level max, adjust compaction
  thresholds, disable agent teams, add background tasks and no-flicker
- main-system-prompt.md: streamline prompt content
- file-manifest.json: update file list for new rules
- claude-code-router.json: add default router configuration

* Update agent definitions: remove worktree isolation, upgrade models

- Remove isolation: worktree from write-capable agents (documenter,
  implementer, migrator, refactorer, test-writer) — run in main worktree
- Upgrade investigator and security-auditor from sonnet to opus

* Update documentation and refactor CLAUDE.md structure

- Move devcontainer guide from CLAUDE.md to AGENTS.md
- CLAUDE.md now uses @AGENTS.md include directive
- docs: update changelog, cli-tools reference, troubleshooting
- docs: remove dashboard references, fix sidebar formatting
- deps: update docs package-lock.json

* Pin agent models to opus-4-5

Update all 9 opus-based agents to use explicit model version opus-4-5.

* Bump version to 2.2.0 with comprehensive changelog

Update changelog with all changes since v2.1.1:
- Claude Code Router and Codex CLI features
- Dashboard feature removed from devcontainer
- 6 new behavioral rules
- Agent model pinning to opus-4-5
- Worktree isolation removed from write-capable agents
- Configuration updates (opus-4-7, reduced thinking tokens, max effort)
- Hook gate path changed to ~/.claude/
- Config directory restructure
- Documentation overhaul

Version updated in package.json and README.md.

* Add effort frontmatter to all agents and skills

Configure Opus 4.7 adaptive thinking via effort: field in frontmatter:
- 19 agents: max (2), xhigh (10), high (4), medium (5)
- 23 skills: xhigh (3), high (10), medium (9), low (1)

Effort levels calibrated by task complexity:
- max: architect, spec-writer (deep reasoning required)
- xhigh: implementer, migrator, refactorer, etc. (code writing)
- high: debug-logs, dependency-analyst, etc. (analysis tasks)
- medium: explorer, bash-exec, etc. (simple operations)
- low: worktree (basic git commands)

* Add session tokens command for thinking token analysis

- New `codeforge session tokens` command analyzes thinking token usage
- Shows exact billed output tokens and visible content breakdown
- Thinking Density table: % turns with thinking, avg chars, session intensity
- Per-session breakdown: turns with thinking, density, avg chars per turn
- Filter by --project, --model, --since, --until
- Output formats: text (colorized) and JSON

Bump CLI version to 0.2.0

* Add oh-my-claude devcontainer feature (disabled)

Add oh-my-claude multi-provider proxy feature for Chinese LLM routing
(Kimi, DeepSeek, Qwen, Zhipu, MiniMax). Feature is disabled pending
refinement — omc install modifies settings.json despite --skip-* flags.

Feature structure:
- devcontainer-feature.json with version, autostart, providerAgentsOnly options
- install.sh with backup/restore approach to protect settings.json
- poststart-hook.sh for auto-starting proxy
- README.md documenting usage and configuration

Supporting changes:
- .secrets.example: Add Chinese LLM provider API keys
- disabled-hooks.json: Add omc memory/preference hooks
- setup-aliases.sh: Add --disallowedTools for omc MCP tools, omc-apply alias
- devcontainer.json: Add Chinese provider secrets (feature commented out)

Note: claude-code-router also commented out pending router consolidation.

* Ignore .research/ directory

* Fix documentation and security issues from PR #71 review

- Fix CLI version in README (0.1.0 → 0.2.0)
- Update 16 path references in AGENTS.md (.codeforge/ → defaults/codeforge/)
- Replace unsafe eval pattern with getent passwd in 9 install scripts

The eval echo "~$USERNAME" pattern could theoretically allow command
injection if username validation were removed in future changes.
Using getent passwd is safer and matches the pattern already used
in codex-cli/install.sh.

Author: AnExiledDev

.gitignore

@@ -20,6 +20,12 @@ coverage/
 *.swp
 *.swo
 *~
+TEMP/
+temp/
+tmp/
+
+# Screenshots
+*.png
 
 # IDE
 .vscode/
@@ -60,10 +66,8 @@ cli/.pytest_cache/
 cli/.ruff_cache/
 .codeforge/data/
 
-# Dashboard-specific
-dashboard/.svelte-kit/
-dashboard/build/
-dashboard/mockups/
+# Dashboard (deprecated — migrating to CodeDirective, a separate repo)
+dashboard/
 
 # Docs-specific
 docs/.astro/
@@ -74,3 +78,4 @@ dev-debug.log
 npm-debug.log*
 yarn-debug.log*
 yarn-error.log*
+.research/

CLAUDE.md

@@ -1,6 +1,6 @@
 # CodeForge Monorepo
 
-This repository contains four packages. Each package manages its own dependencies independently.
+This repository contains three packages. Each package manages its own dependencies independently.
 
 ## Packages
 
@@ -9,7 +9,6 @@ This repository contains four packages. Each package manages its own dependencie
 | `container/` | Node.js | npm |
 | `cli/` | Bun | bun |
 | `docs/` | Node.js | npm |
-| `dashboard/` | Bun | npm (frontend) / bun (server) |
 
 ## Development Rules
 
@@ -27,8 +26,6 @@ Each package has its own `CLAUDE.md` with package-specific development rules:
 - [`container/CLAUDE.md`](container/CLAUDE.md) — changelog, documentation, and configuration rules for the devcontainer package
 - `cli/` — Bun/TypeScript CLI; run `bun test` for tests
 - `docs/` — Astro/Starlight site; run `npm run build` to verify
-- [`dashboard/CLAUDE.md`](dashboard/CLAUDE.md) | [`dashboard/README.md`](dashboard/README.md) — Svelte 5 SPA + Bun backend for session analytics
-
 ### Cross-Package Changes
 
 When a change spans multiple packages, make the changes in a single branch and PR.
@@ -41,21 +38,3 @@ Run tests for each affected package before committing:
 - **Container**: `cd container && npm test`
 - **CLI**: `cd cli && bun test`
 - **Docs**: `cd docs && npm run build`
-- **Dashboard**: `cd dashboard && bun test`
-
-### Dashboard vs CLI
-
-The `dashboard/` and `cli/` packages serve different audiences:
-
-- **CLI** (`codeforge` command) — terminal-first, text/JSON output, scriptable,
-  runs inside or outside the container. Features: session search, task search/list/show,
-  plan search, plugin management, index/config commands.
-- **Dashboard** (Svelte 5 SPA) — visual analytics, charts, expandable detail views,
-  real-time SSE updates. Features: session browsing with conversation replay,
-  task/plan/agent/memory views, project analytics, cost tracking.
-
-When adding a new data view:
-- If it's browsable/visual (tables, charts, detail drill-down) → dashboard
-- If it's scriptable/automatable (piped output, filters, JSON) → CLI
-- If it's both → implement in both, but don't import CLI as a dashboard dependency.
-  Fork patterns instead.

README.md

@@ -6,9 +6,8 @@ Monorepo for CodeForge — an AI-powered development environment for Claude Code
 
 | Package | Description | Version |
 |---------|-------------|---------|
-| [`container/`](container/) | CodeForge DevContainer (`@coredirective/cf-container` on npm) | 2.1.0 |
-| [`cli/`](cli/) | CodeForge CLI (`codeforge-cli`) | 0.1.0 |
-| [`dashboard/`](dashboard/) | Session analytics dashboard (Svelte 5 SPA + Bun backend) | — |
+| [`container/`](container/) | CodeForge DevContainer (`@coredirective/cf-container` on npm) | 2.2.0 |
+| [`cli/`](cli/) | CodeForge CLI (`codeforge-cli`) | 0.2.0 |
 | [`docs/`](docs/) | Documentation site ([codeforge.core-directive.com](https://codeforge.core-directive.com)) | — |
 
 ## Quick Start
@@ -31,9 +30,6 @@ cd container && npm test
 # CLI (Bun)
 cd cli && bun test
 
-# Dashboard (Bun)
-cd dashboard && bun test
-
 # Docs (npm)
 cd docs && npm run build
 ```

cli/CHANGELOG.md

@@ -1,5 +1,22 @@
 # CodeForge CLI Changelog
 
+## v0.2.0 — 2026-04-16
+
+### New Command
+
+- **`codeforge session tokens`** — Analyze thinking token usage across Claude Code sessions
+  - Shows exact billed output tokens (from API) and visible content breakdown (thinking, text, tool chars)
+  - **Thinking Density** table: % of turns with thinking, avg chars per thinking turn, session intensity breakdown (none/light/medium/heavy)
+  - **Per-session breakdown**: turns with thinking, density %, avg chars per thinking turn
+  - Filter by `--project`, `--model`, `--since`, `--until`
+  - Output formats: text (colorized tables) and JSON (`--format json`)
+
+### Purpose
+
+Benchmarks thinking token costs to compare extended thinking behavior across models (e.g., Opus 4.5 vs 4.6). Since `output_tokens` is a combined total (thinking + text + tool_use) with no separate `thinking_tokens` field, this command provides visibility into thinking patterns through empirical measurement of visible content.
+
+---
+
 ## v0.1.0 — 2026-03-14 (Experimental)
 
 ### Package Rename

cli/package.json

@@ -3,7 +3,7 @@
 	"publishConfig": {
 		"access": "public"
 	},
-	"version": "0.1.0",
+	"version": "0.2.0",
 	"description": "CLI for CodeForge development workflows",
 	"keywords": [
 		"codeforge",