feat: add Claude Code skill and Gemini CLI extension

.claude/skills/rpg/SKILL.md

@@ -0,0 +1,137 @@
+---
+name: rpg
+description: 'Build and query semantic code graphs using RPG-Encoder. Use when the user wants to understand codebase structure, search for code by intent, explore dependencies, analyze change impact, or perform semantic lifting.'
+allowed-tools: Bash(rpg-encoder *), Read, Glob, Grep
+---
+
+# RPG-Encoder CLI Skill
+
+You have access to `rpg-encoder`, a CLI tool that builds semantic code graphs (Repository Planning Graphs) from any codebase. Use it to understand code structure, search by intent, trace dependencies, and perform autonomous semantic lifting.
+
+## Quick Reference
+
+### Build a graph (run once per repo)
+
+```bash
+rpg-encoder build
+```
+
+Detects languages automatically, parses all source files, builds structural hierarchy and dependency edges. Creates `.rpg/graph.json`. Use `--force` to rebuild.
+
+### Search for code by intent
+
+```bash
+rpg-encoder search "validate user input"
+rpg-encoder search "database connection" --mode features
+rpg-encoder search "auth" --mode snippets
+rpg-encoder search "parse config" --scope DataProcessing
+```
+
+Modes: `auto` (default, tries both), `features` (semantic intent), `snippets` (name/path matching).
+
+### Fetch entity details
+
+```bash
+rpg-encoder fetch "src/auth.rs:validate_token"
+```
+
+Returns entity type, file location, semantic features, source code, and dependency edges (invokes/invoked-by).
+
+### Explore dependency graph
+
+```bash
+rpg-encoder explore "src/auth.rs:validate_token" --direction both --depth 3
+```
+
+Directions: `up` (what calls it), `down` (what it calls), `both`.
+
+### Show graph statistics
+
+```bash
+rpg-encoder info
+```
+
+Shows entity count, file count, lifting coverage, hierarchy, functional areas, edge counts.
+
+### Autonomous semantic lifting
+
+```bash
+# Estimate cost first
+rpg-encoder lift --dry-run
+
+# Lift with Anthropic Haiku (default, cheapest)
+rpg-encoder lift --provider anthropic
+
+# Lift with OpenAI GPT-4o-mini
+rpg-encoder lift --provider openai --model gpt-4o-mini
+
+# Lift specific scope
+rpg-encoder lift --scope "src/auth/**"
+```
+
+Performs full autonomous lifting: auto-lift trivial entities, LLM-lift the rest, synthesize file features, discover domains, and assign hierarchy. Saves after each batch for crash recovery.
+
+API key: pass `--api-key KEY` or set `ANTHROPIC_API_KEY` / `OPENAI_API_KEY` env var.
+
+### Incremental update (after code changes)
+
+```bash
+rpg-encoder update
+rpg-encoder diff  # dry-run: see what would change
+```
+
+### Export graph
+
+```bash
+rpg-encoder export --format dot      # Graphviz DOT
+rpg-encoder export --format mermaid  # Mermaid flowchart
+```
+
+### Validate graph integrity
+
+```bash
+rpg-encoder validate
+```
+
+### Git hook (auto-update on commit)
+
+```bash
+rpg-encoder hook install
+rpg-encoder hook uninstall
+```
+
+## Workflow Patterns
+
+### Understanding a new codebase
+
+1. `rpg-encoder build` to index
+2. `rpg-encoder info` to see structure
+3. `rpg-encoder search "main entry point"` to find key entities
+4. `rpg-encoder explore ENTITY --direction down` to trace call chains
+
+### Finding code by what it does
+
+Use `--mode features` with natural language describing behavior:
+```bash
+rpg-encoder search "handle authentication" --mode features
+rpg-encoder search "serialize data to JSON" --mode features
+```
+
+### Change impact analysis
+
+1. Find the entity: `rpg-encoder search "function_name"`
+2. See what depends on it: `rpg-encoder explore ENTITY --direction up --depth 3`
+
+### Full semantic analysis
+
+1. `rpg-encoder build` (if not done)
+2. `rpg-encoder lift --dry-run` (check cost)
+3. `rpg-encoder lift` (run lifting)
+4. `rpg-encoder info` (verify coverage)
+
+## Notes
+
+- The CLI works from the project root directory. Use `-p /path/to/project` to target a different directory.
+- Graph is saved at `.rpg/graph.json` — add `.rpg/` to `.gitignore` or commit it for team sharing.
+- Lifting is resumable: if interrupted, re-running `rpg-encoder lift` continues from where it stopped.
+- For richer interactive navigation (context packs, impact radius, path finding), use the MCP server instead.

.gemini/extensions/rpg/CONTEXT.md

@@ -0,0 +1,20 @@
+# RPG-Encoder
+
+RPG-Encoder builds semantic code graphs (Repository Planning Graphs) for AI-assisted code understanding.
+
+## CLI Commands
+
+- `rpg-encoder build` — Index codebase, build graph (run once)
+- `rpg-encoder search "query"` — Search entities by intent or name
+- `rpg-encoder fetch ENTITY_ID` — Get entity details and source
+- `rpg-encoder explore ENTITY_ID` — Trace dependency chains
+- `rpg-encoder info` — Show graph statistics
+- `rpg-encoder lift` — Autonomous LLM-driven semantic lifting
+- `rpg-encoder lift --dry-run` — Estimate lifting cost
+- `rpg-encoder update` — Incremental update after code changes
+- `rpg-encoder export --format dot|mermaid` — Export graph
+- `rpg-encoder validate` — Check graph integrity
+
+## MCP Tools (via extension)
+
+The extension also provides an MCP server with rich navigation tools: `search_node`, `fetch_node`, `explore_rpg`, `context_pack`, `impact_radius`, `plan_change`, `find_paths`, and the full lifting protocol.

.gemini/extensions/rpg/commands/rpg-build.toml

@@ -0,0 +1,13 @@
+[command]
+description = "Build a semantic code graph (RPG) from the codebase"
+
+[command.prompt]
+text = """
+Build a Repository Planning Graph for this codebase using the rpg-encoder CLI.
+
+Run: `rpg-encoder build`
+
+This indexes all source files, detects languages, extracts entities, builds the dependency graph and structural hierarchy. The graph is saved to `.rpg/graph.json`.
+
+If the graph already exists, use `--force` to rebuild. After building, run `rpg-encoder info` to show the result.
+"""

.gemini/extensions/rpg/commands/rpg-lift.toml

@@ -0,0 +1,22 @@
+[command]
+description = "Run autonomous LLM-driven semantic lifting on the codebase"
+
+[command.prompt]
+text = """
+Perform autonomous semantic lifting on this codebase using cheap LLM APIs.
+
+First, estimate the cost: `rpg-encoder lift --dry-run`
+
+Show the estimate to the user. If they approve, run the full lift:
+`rpg-encoder lift --provider anthropic`
+
+The lift pipeline:
+1. Auto-lifts trivial entities (getters, setters, constructors) without LLM calls
+2. Sends remaining entities to the LLM in batches for semantic feature extraction
+3. Synthesizes file-level features
+4. Discovers functional domains and assigns 3-level hierarchy
+
+Progress is saved after each batch. If interrupted, re-running continues from where it stopped.
+
+After completion, run `rpg-encoder info` to show the updated graph statistics.
+"""

.gemini/extensions/rpg/commands/rpg-search.toml

@@ -0,0 +1,15 @@
+[command]
+description = "Search for code entities by intent or keywords"
+
+[command.prompt]
+text = """
+Search the codebase semantic graph for entities matching the user's query.
+
+Run: `rpg-encoder search "$PROMPT" --mode auto`
+
+Use `--mode features` for intent-based search (e.g., "validate user input", "handle HTTP errors").
+Use `--mode snippets` for name/path matching (e.g., "auth", "parse_config").
+Use `--scope AREA` to restrict to a functional area.
+
+Show the results and explain what each entity does based on its features and location.
+"""

.gemini/extensions/rpg/gemini-extension.json

@@ -0,0 +1,12 @@
+{
+  "name": "rpg-encoder",
+  "version": "0.7.0",
+  "description": "Build and query semantic code graphs (Repository Planning Graphs) for AI-assisted code understanding. Provides entity search, dependency exploration, and autonomous LLM-driven semantic lifting.",
+  "mcpServers": {
+    "rpg-encoder": {
+      "command": "npx",
+      "args": ["-y", "rpg-encoder@latest", "mcp"]
+    }
+  },
+  "contextFileName": "CONTEXT.md"
+}