By jobrien127
Meta-plugin that injects project state into every session and synthesizes session transcripts into a knowledge base on session end. Includes skills for routing, reasoning, verification, and context compression.
npx claudepluginhub jobrien127/plugin-marketplace --plugin whetstoneUse when conversation exceeds 10 turns or context feels large. Produces a compact # Context Reset block preserving task, completed work, remaining steps, constraints, and key decisions. Invoke before switching tasks or when earlier context may be lost.
Improve a whetstone skill based on an improvement spec. Analyzes the spec, understands the gap, makes a targeted edit to the skill, and tests the fix.
Analyze KB learnings and test scores to identify the highest-value skill improvement. Spawns the whetstone:optimizer agent, which writes an improvement spec consumed by /whetstone:improve.
Use before starting any task requiring more than 2 tool calls. Classifies task complexity, domain, and clarity, then recommends direct answer, skill chain, decomposition, or agent delegation with estimated cost.
Initialize whetstone KB directories and state files. Run once after plugin install. Creates ~/.whetstone/ structure and initial config.
Manually trigger session synthesis. Synthesis now runs automatically on session-end, but use this to re-synthesize a specific session with a different model.
Use before committing to an implementation when confidence is below 0.8, when facing an architectural decision, or when the problem is novel. Produces a CONFIDENCE-scored DECISION with explicit reasoning trace.
Use before reporting any task complete. Checks completeness against the original request, correctness, edge cases, and regressions. Returns VERDICT: SHIP, REVISE, or ESCALATE with specific gaps.
A local marketplace of Claude Code plugins — self-contained packages that extend Claude with hooks, skills, agents, and MCP servers.
v1.4.1Meta-plugin that injects project state into every session and synthesizes session transcripts into a persistent knowledge base. The backbone for self-improving Claude Code setups.
Skills: /whetstone:route, /whetstone:think, /whetstone:verify, /whetstone:compress, /whetstone:synthesizer, /whetstone:improve, /whetstone:setup
Auto-behaviors:
v0.1.0Routes tasks to the appropriate Claude model tier (Haiku / Sonnet / Opus) based on task complexity, reducing cost without sacrificing quality.
Skills: /dispatch:route, /dispatch:haiku, /dispatch:sonnet, /dispatch:opus
Interactive wizard for managing project-level settings.local.json permissions. Detects project type, suggests safe permission patterns, and reduces manual approval friction.
Command: /permissions
Features: Smart project-type detection, categorized permission safety levels, proactive suggestions when starting new projects.
Transparently converts structured data in prompts to TOON format for token-efficient LLM processing, then converts responses back to JSON for display. Saves 30–60% tokens on structured data.
Extraction methods: LLM (Claude API), Markdown tables, key-value regex — with graceful fallback.
v0.1.0MCP server and hook-driven automation for structured problem-solving workflows. Decomposes problems, specifies requirements, generates solutions, validates output, and resumes context across sessions.
Agents: decomposer, specifier, generator, validator, calibrator
Hooks: SessionStart (resume context), PostToolUse (validate stage), PreCompact/PostCompact (save/restore state)
Requires: Python ≥ 3.12, uv
v0.1.0Automated wiki maintenance for this plugin marketplace. Ingests plugin source files, lints wiki docs, and keeps the knowledge base in sync.
Agents: ingest-agent, lint-agent, remediate-agent, obsidian-agent
Each plugin is self-contained. Install by pointing Claude Code at the plugin directory:
# Load a single plugin
claude --plugin-dir /path/to/plugin-marketplace/whetstone
# Or copy to your Claude plugins directory
cp -r whetstone ~/.claude/plugins/
Then enable in ~/.claude/settings.json:
{
"enabledPlugins": {
"whetstone": true
}
}
Each plugin follows the Claude Code plugin spec:
<plugin-name>/
├── .claude-plugin/
│ └── plugin.json # Manifest: name, version, description, hooks
├── skills/ # Slash commands (/plugin:skill)
├── agents/ # Subagent prompt definitions
├── hooks/ # Lifecycle hook scripts
├── commands/ # Top-level slash commands
└── README.md
MIT
Ultra-compressed communication mode. Cuts ~75% of tokens while keeping full technical accuracy by speaking like a caveman.
Matches all tools
Hooks run on every tool call, not just specific ones
Executes bash commands
Hook triggers when Bash tool is used
Modifies files
Hook triggers on file write and edit operations
Comprehensive UI/UX design plugin for mobile (iOS, Android, React Native) and web applications with design systems, accessibility, and modern patterns
Intelligent prompt optimization using skill-based architecture. Enriches vague prompts with research-based clarifying questions before Claude Code executes them
Persistent memory system for Claude Code - seamlessly preserve context across sessions
Memory compression system for Claude Code - persist context across sessions
Team-oriented workflow plugin with role agents, 27 specialist agents, ECC-inspired commands, layered rules, and hooks skeleton.
Modifies files
Hook triggers on file write and edit operations