By kyuna0312
Claude Code plugin combining token-waste reduction (skills, hooks, status line) with DB-backed project scaffolding, changelog tracking, and template back-mapping via MCP.
This skill should be used when the user says "auto compact", "enable auto compact", "compact on context full", "automatic context compression", "prevent context limit errors", "set up auto compaction", or "context keeps hitting limit".
This skill should be used when the user says "check claude.md size", "is my claude.md too big", "claude.md token cost", "how big is claude.md", or when triggered automatically on session start to warn about oversized claude.md files.
This skill should be used when the user says "startup hook error", "hook not working", "debug hooks", "fix hook error", "SessionStart error", "hook script failing", "node:internal/modules error", "diagnose hook", or sees hook-related errors in Claude Code.
This skill should be used when the user says "estimate token usage", "how many tokens am I using", "token cost of my setup", "audit token budget", "what's eating my context", "token usage breakdown", or "how expensive is my session".
This skill should be used when the user asks to "show recent changes", "show forge changelog", "show changelog for <project>", "what did I add this week", "what files did I touch today", "what packages keep recurring", "what drifts from the template", "review template suggestions", "show pending forge suggestions", "sync forge templates", "apply template suggestion", or "compute back-mapping suggestions". Provides read, drift-discovery, and template back-mapping workflows that wrap the `forge-db` MCP tools (`get_changelog`, `compute_suggestions`, `apply_suggestion`).
Admin access level
Server config contains admin-level keywords
Modifies files
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
Hook triggers on file write and edit operations
Hook triggers on file write and edit operations
Uses power tools
Uses Bash, Write, or Edit tools
Uses power tools
Uses Bash, Write, or Edit tools
A Claude Code plugin that combines two things:
/scaffold, /changelog, /sync-template slash commands plus an MCP server (forge-db) that stores templates, file content, and pinned dependency versions in Postgres so the model never invents them.| Skill | What it does |
|---|---|
optimize-claudemd | Compresses bloated CLAUDE.md — cuts 40–80% of tokens loaded every session |
low-token-mode | Switches Claude to terse response style — cuts reply tokens by 30–50% |
reset-context | Safely resets context window when near full — prevents hard stops |
tune-settings | Diffs and applies token-saving settings (autoLoadMemory, autoLoadSkills, etc.) |
manage-skills | Audits loaded skills, disables unused ones — reduces session overhead |
project-isolation | Scopes skills and hooks to current project only |
estimate-tokens | Estimates tokens in any file before loading it |
auto-compact | Configures compactOnContextFull — auto-compacts instead of stopping |
settings-diff | Shows before/after diff before writing any settings change |
check-claudemd-size | Reports CLAUDE.md word/token count with color-coded warnings |
token-statusline | Adds live context bar to Claude Code status line |
debug-hooks | Diagnoses broken hook configurations with detailed validation output |
task-brain-lite | Decomposes complex tasks into prioritized, dependency-aware subtasks |
llm-wiki | Builds a persistent wiki Claude references instead of re-reading raw docs — up to 96% token savings on repeated knowledge |
forge-changelog | Reads the forge changelog and runs the template back-mapping loop (wraps forge-db MCP tools get_changelog, compute_suggestions, apply_suggestion) |
Agent: hook-error-fixer — diagnoses and auto-fixes broken hook configurations.
Hook: Session-start script that warns when CLAUDE.md exceeds size thresholds.
Status line:
ctx [████████░░] 82% │ md:~650t
Three slash commands backed by the forge-db MCP server:
| Command | What it does |
|---|---|
/scaffold | Create a project from a template stored in Postgres |
/changelog | Show recorded file/dependency changes for a project |
/sync-template | Review recurring manual additions and fold them back in |
Why facts live in Postgres: templates, file contents, and exact dependency versions are read verbatim from the DB through MCP tools. The model copies — it does not invent template names or guess versions — and scaffolds run a real typecheck/build before being declared good. Retrieval + validation instead of fine-tuning.
A PostToolUse hook (record-change.mjs) appends every Write / Edit to the changelogs table; /sync-template later analyses those rows and suggests template improvements.
# 1. Install MCP server deps
cd mcp && npm install && cd ..
# 2. Point at your remote Postgres
export FORGE_DATABASE_URL="[REDACTED:DB Connection String]
# 3. Create the schema (and an example template to test with)
psql "$FORGE_DATABASE_URL" -f mcp/db/schema.sql
psql "$FORGE_DATABASE_URL" -f mcp/db/seed-example.sql
.mcp.json reads ${FORGE_DATABASE_URL} from your environment, so make sure that variable is exported in the shell where you launch Claude Code. Without it, the forge half is inert — the token-saver half keeps working.
forge-db)list_templates, get_template, register_project, record_change, get_changelog, compute_suggestions, apply_suggestion.
refreshInterval support)python3 — used by the token status line script and hook validationnode ≥ 18 — runs the MCP server and the record-change hook$FORGE_DATABASE_URL (forge half only — token-saver half does not need it)git clone https://github.com/kyuna0312/context_forge.git ~/.claude/plugins/context_forge
git clone https://github.com/kyuna0312/context_forge.git ~/context_forge
claude --plugin-dir ~/context_forge
git clone https://github.com/kyuna0312/context_forge.git ~/context_forge
bash ~/context_forge/scripts/install.sh
The install script symlinks the plugin into ~/.claude/plugins/context_forge.
claude --plugin-dir ~/Desktop/context_forge
npx claudepluginhub kyuna0312/context_forge --plugin context_forgeOpen-source, local-first Claude Code plugin for token reduction, context compression, and cost optimization using hybrid RAG retrieval (BM25 + vector search), reranking, AST-aware chunking, and compact context packets.
Complete AI coding workflow system. Self-correcting memory + persistent FTS5-indexed research wikis + auto-research loop + multi-LLM council on a single SQLite store. 33 skills, 8 agents, 22 commands, 37 hook scripts across 24 events. Cross-agent via SkillKit.
Intelligent draw.io diagramming plugin with AI-powered diagram generation, multi-platform embedding (GitHub, Confluence, Azure DevOps, Notion, Teams, Harness), conditional formatting, live data binding, and MCP server integration for programmatic diagram creation and management.
Give your AI a memory — mine projects and conversations into a searchable palace. 33 MCP tools, auto-save hooks, and guided setup.
MCP server that saves 98% of your context window with session continuity. Sandboxed code execution in 11 languages, FTS5 knowledge base with BM25 ranking, and automatic state restore across compactions.
Complete creative writing suite with 10 specialized agents covering the full writing process: research gathering, character development, story architecture, world-building, dialogue coaching, editing/review, outlining, content strategy, believability auditing, and prose style/voice analysis. Includes genre-specific guides, templates, and quality checklists.