CHANGELOG.md

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[0.8.3] - 2026-04-14

Added

• lifting_status now tracks stale-feature drift across calls. A persistent per-server set records entities whose source was modified after they were lifted; the dashboard reports stale_features: N entities modified since last lift and the NEXT STEP state machine prompts re-lift even when coverage is 100%.
• get_entities_for_lifting(scope="*") now returns stale entities alongside unlifted ones, so a single call covers both "never lifted" and "lifted-but-outdated" work.
• lifting_status emits a sub-agent dispatch recommendation when the remaining work (unlifted + stale) is ≥100 entities. The response contains LOOP / DISPATCH / FALLBACK blocks so callers delegate directly without first loading a batch of source into their own context.
• get_entities_for_lifting batch-0 emits a one-line dispatch hint when ≥10 token-aware batches are queued, pointing back to lifting_status for the full recommendation.
• submit_lift_results NEXT action is now scale-aware — when remaining work ≥100 entities, it redirects the caller to lifting_status instead of encouraging another foreground batch.
• build_rpg response now emits an action-oriented NEXT STEP directing the agent to lift immediately. Small scope lifts inline; large scope dispatches a sub-agent with the LOOP pattern embedded.
• New USE RPG FIRST top section in server_instructions.md with a mapping table from shell patterns (grep -r, cat, find, wc -l, chained greps) to the RPG tool that replaces them.
• New DRIFT MAINTENANCE section in server_instructions.md explaining the three auto-sync notice variants and framing re-lift as part of definition-of-done for any task that wrote code.
• Tool descriptions for search_node, fetch_node, explore_rpg, rpg_info, semantic_snapshot, context_pack, impact_radius, plan_change, analyze_health, detect_cycles, find_paths, and slice_between now open with a PREFER THIS OVER ... marker naming the shell command or workflow they replace.
• .claude/skills/rpg/SKILL.md and README.md carry the same RPG-first mapping table as the server prompt.
• Crate-visible LARGE_SCOPE_ENTITIES (100) and LARGE_SCOPE_BATCHES (10) constants replace duplicated magic numbers across server.rs/tools.rs with doc comments describing the heuristic-vs-authoritative relationship.
• Canonical lock-order invariant documented on the RpgServer struct so reviewers don't have to re-derive it from scattered call sites.
• reload_config_with_warning helper on RpgServer that distinguishes missing .rpg/config.toml (silent default) from malformed TOML (stderr warning, keeps previous in-memory config).

Changed

• lifting_status NEXT STEP is runtime-neutral. No specific runtime's dispatch syntax appears in the response; callers use whatever sub-agent or cheaper-model mechanism their runtime exposes. Explicit fallbacks: scoped lifting for callers with no delegation mechanism, and rpg-encoder lift --provider anthropic|openai for callers with an API key and no sub-agent support.
• Batch-size estimates in NEXT STEP messages read from the live encoding.max_batch_tokens config instead of a hard-coded ~12K figure, so the estimate scales when the user overrides the budget.
• NEXT STEP: remains a single parseable line; dispatch detail is emitted in labeled blocks immediately below.
• Auto-sync notice now prescribes a verb: it distinguishes per-update delta from pre-existing backlog and separates new-entity drift from stale-feature drift, so an agent that commits code and sees the notice is told to re-lift rather than informed of a count.
• CLI fallback in large-scope guidance is gated to cases with actual unlifted work, with a note that rpg-encoder lift does not re-lift stale entities (it filters to entities with no features).
• get_routing_candidates response header no longer includes the graph revision hash — it moved to the NEXT_ACTION block at the bottom. Keeps the stable preamble (instructions + entity table) cache-eligible while still surfacing the revision where the agent needs to read it back.
• server_instructions.md LIFTING FLOW sub-section rewritten and shortened.
• set_project_root on project switch now loads the new project's .rpg/config.toml independently; on parse failure it falls back to RpgConfig::default() rather than preserving the previous project's config. Same-project reload_rpg preserves the previous config on parse failure (different flow, different correctness requirement).

Fixed

• lifting_status previously reported Graph is complete as soon as every entity had some features, ignoring stale features from modified sources. The state machine now considers remaining + stale_features combined.
• get_entities_for_lifting(scope="*") previously returned zero entities when all drift was stale (features present, sources modified) because resolve_scope filters to entities with no features. It now augments the resolved scope with tracked stale entities and routes them through the LLM loop.
• Auto-sync notice previously conflated per-update delta with global backlog, so a one-line edit on a partially-lifted repo could claim "50 new entities unlifted" when only 1 was actually new.
• finalize_lifting fallback guidance previously said to call it after each scoped subtree. That auto-routes pending entities against incomplete signals and locks the hierarchy early. Guidance now says to call finalize_lifting once after all scopes complete.
• rpg-encoder lift --provider ... (CLI fallback) left the MCP server on a stale in-memory graph. All docs that mention the CLI fallback now specify that the caller must call reload_rpg afterward.
• set_project_root and reload_rpg previously used unwrap_or_default() on config loads, collapsing missing-config and malformed-TOML into identical silent behavior.
• set_project_root failed to refresh self.config on project switch; the server kept serving the previous project's encoding settings.
• lifting_status large-scope recommendation previously ran off raw unlifted count, before auto-lift had reduced the set. On repos full of trivial entities (getters, setters, constructors) it could recommend delegation for ~0 LLM calls. The large-scope branch now signals likely-large and defers the authoritative check to the post-auto-lift batch count in get_entities_for_lifting.
• rpg_info error wording ("No RPG found") was miscited as a friendly status string; corrected to "any RPG tool returns 'No RPG found'".
• build_rpg NEXT STEP and lifted: X/Y header previously counted Module entities against the unlifted total, while lifting_status and get_entities_for_lifting exclude them. The two could disagree by hundreds of entities on large codebases, tripping the delegation threshold in build_rpg when lifting_status would still recommend foreground lifting. Both paths now use lifting_coverage() (non-module) for the count, and the header reads liftable_entities: X/Y.
• submit_lift_results previously emitted DONE as soon as coverage reached 100%, which could terminate a stale-only re-lift loop after batch 1 while later batches were still queued. The NEXT/DONE branch now counts unlifted + stale remaining.
• update_rpg now feeds summary.modified_entity_ids into the stale-tracking set so its needs_relift: N reply aligns with what lifting_status and get_entities_for_lifting(scope="*") report.
• Server startup auto-update now feeds modified_entity_ids into the stale-tracking set and seeds the auto-sync changeset hash for a clean workdir. Previously modifications between the last lift and a session restart silently dropped off the dashboard.
• auto_lift on a non-* scope now drains stale entries for every in-scope entity. The pipeline freshens features for each regardless of existing state, so stale markers for those IDs are invalid after the call; the unconditional drain also handles the identical-features case where a cosmetic edit re-lifts to the same output.
• Auto-lifted features for entities previously flagged stale now drain the stale-tracking set inline in get_entities_for_lifting, so the count doesn't inflate when the auto-lifter writes fresh features directly.
• reload_rpg now prunes the stale-tracking set against the newly loaded graph rather than clearing it wholesale. The CLI / isolated sub-agent re-lift flow only refreshes entities with no features — stale entities survive it, so clearing would let lifting_status report 100% coverage while re-lift work remained.
• reload_rpg drift-tracking reset now sits on the success path, after storage::load returns Ok. Transient read errors no longer wipe the backlog while leaving the previous graph in memory.
• build_rpg now prunes the stale-tracking set against the newly built graph so dead entity IDs don't accumulate across rebuilds.
• build_semantic_hierarchy sharded init no longer acquires hierarchy_session.write() before graph.read(). The original order formed a deadlock cycle with update_rpg's graph-then-session order under concurrent scheduling. The init path now collapses decision + snapshot into a single hierarchy_session.write() held under graph.read() and packages the work into an Action enum so there's no peek-then-trust.
• build_batch_0_domain_discovery and build_cluster_batch now take &RPGraph and (where applicable) clusters as parameters instead of re-reading self.graph / self.hierarchy_session. Closes two TOCTOU windows: a session-clear race where a concurrent build_rpg/update_rpg could panic on session.as_mut().unwrap(), and a graph-replace race where a concurrent set_project_root could panic on graph.as_ref().unwrap().
• set_project_root tool description is no longer Claude-Code-specific in its example; reads runtime-neutral.
• get_entities_for_lifting batch-0 dispatch NOTE no longer references "batches 2..N" (off-by-one against the 0-based batch_index parameter). Reads "do not request further batches in this context".

[0.8.2] - 2026-04-14

Added

• set_project_root MCP tool — switch the active project root at runtime without restarting the session. Loads the new root's .rpg/graph.json if present, resets lifting/hierarchy sessions, auto-sync markers, and pending routing state. Tilde-expands and canonicalizes the supplied path. Fixes the common case where a session is launched from $HOME but the user wants to work on ~/some-project — previously the server's project root was locked to the launch directory.
• MCP tool count: 27 → 28.

Changed

• RpgServer::project_root is now an async accessor backed by Arc<RwLock<PathBuf>> (previously a static PathBuf field). All tool handlers acquire a snapshot at call time, so each invocation reads whatever set_project_root most recently set.
• get_config_blocking renamed to load_config and made async.
• staleness_detail made async (needed to acquire the new project-root lock).

[0.8.1] - 2026-04-14

Fixed

• CRITICAL: Auto-sync on workdir changes was a no-op (caught by Codex audit) — v0.8.0 detected workdir changes and computed a hash, but the underlying run_update only diffed base_commit..HEAD, so uncommitted edits were silently ignored. The hash was then cached as "synced", masking the staleness from later queries. This made the headline v0.8.0 feature (workdir auto-sync) effectively non-functional.
• Auto-sync error path no longer caches markers — transient failures no longer leave the server silently stale. The next query retries.
• Revert detection — when a previously-dirty file returns to its HEAD state, the graph is now restored. Previously, the entity additions from the dirty version persisted forever.
• CLI update now defaults to workdir-aware — matches the MCP server. Pass --since <commit> for commit-range diffing as before.
• README said "six crates" while listing seven. Now says seven.
• tools.rs module comment said "17 tools" — actually 27.

Added

• run_update_workdir in rpg-encoder::evolution — public API that applies the working-tree diff (committed + staged + unstaged) instead of the committed-only diff.
• run_update_from_changes in rpg-encoder::evolution — public API that applies a caller-supplied Vec<FileChange>. Used by the MCP server to compose workdir changes with revert-detection re-parses.

[0.8.0] - 2026-04-13

Added

• semantic_snapshot MCP tool — Whole-repo semantic understanding in one call (~25K tokens for 1000 entities). Includes hot spots: top-10 most-connected entities as architectural backbone.
• auto_lift MCP tool — Autonomous lifting via external LLM API (Anthropic Haiku, OpenAI GPT-4o-mini, OpenRouter, Google Gemini). Supports api_key_env parameter to resolve keys from environment variables (safer than raw keys in tool calls).
• analyze_health MCP tool — Code health metrics: coupling, instability, god objects, clone detection.
• detect_cycles MCP tool — Circular dependency detection.
• Auto-staleness resolution — Server auto-syncs graph when git HEAD moves. Navigation tools no longer show passive stale warnings, they auto-update.
• MCP tool annotations — read_only_hint, destructive_hint, open_world_hint on all tools per MCP 2025-03-26 spec.
• tool_list_changed capability enabled for future progressive tool loading.
• Claude Code hooks — PreToolUse on Write/Edit (auto-context injection), PostToolUse on Bash (auto-update after git commits).
• MCP tool count: 23 → 27

Fixed

• Deadlock in auto_sync_if_stale() error path — write lock held while calling read-lock method (#75)
• auto_lift graph loss on pipeline error — graph now stays in write lock via block_in_place (#75, #77)
• API key exposure in Debug derive — manual Debug impl redacts api_key (#75)
• Concurrent auto_lift guard — lift_in_progress AtomicBool rejects parallel calls (#76)
• Cost estimate overcounting — Review-confidence entities no longer counted as LLM-needed (#76)
• Pre-commit hook supply-chain risk — removed unpinned npx -y fallback (#77)
• Autonomous lift name collision — tries qualified Class::method names when bare name misses (#77)
• Snapshot dep skeleton ambiguity — uses qualified names instead of bare names (#77)
• Shell hook portability — replaced GNU grep -P with POSIX sed (#77)
• Root scope handling in shared search — scope="." and scope="" now behave as unscoped (#70)

Removed

• Unused kodama dependency

Changed

• server.json version updated to 0.7.0 (was stuck at 0.6.7)
• CONTRIBUTING.md corrected: tools are in tools.rs not main.rs
• README rewritten: leads with value proposition, credits inspirations (RPG paper, GitNexus, Serena)
• Dead code removed from rpg-lift progress.rs

[0.7.0] - 2026-04-08

Added

• Claude Code skill (.claude/skills/rpg/SKILL.md) and Gemini CLI extension (#68)
• README updated to cite arXiv paper directly

Changed

• Entity Signatures carried over from v0.6.0
• MCP tool count: 23 (unchanged)

[0.6.2] - 2026-02-21

Fixed

• Auto-migrate existing Windows graphs — When loading graphs created on Windows (schema <2.2.0), all backslash entity IDs are automatically normalized to forward slashes across entities, file_index, edges, and hierarchy. All lifted semantic features and data are fully preserved. Schema version bumped to 2.2.0.

[0.6.1] - 2026-02-21

Fixed

• Windows path format mismatch in entity IDs (#49) — Path::display() produces backslashes on Windows, causing key mismatches when MCP tools receive forward-slash keys from users/LLMs. Added normalize_path() utility in rpg-core::graph and applied it at all ID-generating call sites: RawEntity::id(), resolve_dependencies(), apply_renames(), and create_module_entities(). No-op on Unix. Closes #49.
• Pre-existing clippy inefficient_to_string warnings in critic.rs, hierarchy.rs, and evolution_test.rs.

[0.6.0] - 2026-02-13

Added

• Entity Signatures — Typed function/method signatures extracted from AST (parameters with type annotations, return types). Stored on Entity.signature.
• DataFlow Edges — New EdgeKind::DataFlow for parameter passing and return value flow between entities. Adds data lineage tracking to the dependency graph.
• Area Connectivity — rpg_info now reports inter-area dependency counts, showing which functional areas are most tightly coupled.
• MCP tool count: 23

Changed

• Schema version bumped for signature and data flow support

[0.5.0] - 2026-02-13

Added

• Protocol Deduplication — Version references save 10K-40K tokens per lifting session
  • get_files_for_synthesis uses version references after batch 0: [RPG Protocol: file_synthesis v<hash>]
  • SHA256-based prompt versioning in MCP server (PromptVersions struct)
• Graph Reasoning Tools — Two new MCP tools for dependency path analysis
  • find_paths: K-shortest paths between entities using Yen's algorithm (returns paths of varying lengths)
  • slice_between: Extract minimal connecting subgraph (Steiner tree with strict edge tracking)
  • Eliminates 5-10 manual tool calls for path queries
• Diff-Aware Search — search_node accepts since_commit parameter for PR review workflows
  • Proximity-based ranking: 3x boost for changed entities, 2x for 1-hop neighbors, 1.5x for 2-hop
  • Automatically maps git changes to entity IDs via rpg_encoder::evolution::detect_changes
  • Computes dependency proximity tiers using BFS traversal
  • Boost applied before truncation to ensure changed entities can rank into results
  • 50-70% fewer irrelevant results projected in PR review tasks
• Sharded Hierarchy Foundation — Clustering infrastructure for repos >100 files
  • File clustering with deterministic batching (target: 70 files per cluster)
  • Representative sampling for domain discovery
  • Balance clusters to maintain manageable batch sizes
  • FULL MCP INTEGRATION COMPLETE: Two-phase batched workflow (domain discovery → file assignment per cluster)

Changed

• search_node MCP tool supports diff-aware ranking via since_commit parameter
• SearchParams extended with diff_context field for proximity boosting
• Diff boost now applied before truncation (expands search limit 10x when diff_context present)
• find_paths defaults to max_hops=5 (use -1 for unlimited)
• File clustering uses sorted iteration for deterministic batch assignments

Performance

• Protocol deduplication: ~75 tokens saved per batch after batch 0 (10K-40K total over 20-batch session)
• Graph reasoning tools: Single-call path queries vs 5-10 manual explore/fetch calls
• Diff-aware search: Reduces noise in PR review workflows

Technical

• Added dependencies: sha2 = "0.10" (rpg-mcp), kodama = "0.2" (rpg-encoder, currently unused)
• New modules:
  • rpg-nav/src/diff.rs (189 lines, 5 tests): Proximity computation and boost application
  • rpg-nav/src/paths.rs (290 lines, 7 tests): K-shortest paths with Yen's algorithm
  • rpg-nav/src/slice.rs (300 lines, 8 tests): Minimal subgraph extraction with strict edge tracking
• Extended rpg-encoder/src/hierarchy.rs (+210 lines, 4 tests): Clustering and batching functions
• MCP tool count: 21 → 23 (new: find_paths, slice_between)
• Sharded hierarchy: Session management, automatic clustering, batched domain discovery and file assignment
• All tests passing (466+ tests across workspace)
• Clustering simplified from HAC to deterministic batching (kodama API complexity deferred)

[0.4.0] - 2026-02-13

Added

• Incremental Embeddings — Per-entity fingerprint sync replaces full embedding rebuild. Only re-embeds entities whose features changed. Stored in embeddings.meta.json alongside existing binary format.
• Confidence-Gated Auto-Lift — Structural signal analysis (branches, loops, calls, early returns) in new signals.rs module. Three confidence buckets:
  • Accept: apply features silently (simple getters, setters, constructors)
  • Review: show in batch 0 for LLM verification (moderate complexity)
  • Reject: needs full LLM analysis (high complexity) TOML rules extended with max_branches, max_loops, max_calls structural gates.
• Lift Quality Critic — Non-blocking feedback on submit_lift_results. Checks for vague verbs ("handle", "process"), implementation details ("loop", "iterate"), too-short/too-long features, and duplicates. Features are always applied; warnings help the LLM self-correct.
• plan_change MCP tool — Answers "what existing code needs to change for goal X, and in what order?" Orchestrates search + impact_radius + topological sort + test coverage detection. Returns dependency-safe modification order with blast radius analysis.
• feature_source provenance — Option<String> field on Entity tracks feature origin ("auto", "llm", "synthesized"). Backward-compatible via serde(default).
• MCP tool count: 20 → 21 (new: plan_change)

[0.3.0] - 2026-02-12

Added

• Language-Universal Auto-Lift — 134 TOML-driven auto-lift rules across 13 languages with acronym-aware field normalization (getHTTPClient → return http client) (#40)
• LLM Performance Optimizations — context_pack super-tool (search→fetch→explore in 1 call), impact_radius BFS reachability, dependency context in lifting batches, auto-lift trivial entities (≤3 lines, getter/setter/new patterns) (#38)
• Preserve semantic features on build_rpg rebuild and prune .rpgignore files (#36)
• MCP tool count: 17 → 20 (new: context_pack, impact_radius, reconstruct_plan)

Fixed

• Scope auto-lift rules by language to prevent cross-language collisions (#42)

[0.2.0] - 2026-02-11

Added

• Reconstruction Scheduler — reconstruct_plan builds dependency-safe execution batches for guided code reconstruction workflows (#34)
• Validation improvements and documentation fixes (#34)

Changed

• Refactored rpg-mcp — Split monolithic tools.rs into focused modules: params.rs, types.rs, helpers.rs, server.rs
• Test count: 379 → 446+

[0.1.9] - 2026-02-09

Added

• LLM-Based Semantic Routing (Algorithm 4) — Two new MCP tools (get_routing_candidates, submit_routing_decisions) let the connected agent perform semantic hierarchy placement decisions. Entities are routed via LLM judgment rather than Jaccard similarity alone. Persisted pending state (.rpg/pending_routing.json) with graph revision protection for crash safety and stale-decision rejection.
• Three-Zone Drift Judgment (Algorithm 3) — Configurable drift thresholds split re-lifted entities into three zones:
  • drift < 0.3 (ignore): minor edit, in-place update
  • 0.3 <= drift <= 0.7 (borderline): surfaced for agent review via routing candidates
  • drift > 0.7 (auto-route): automatically queued for re-routing New config options: drift_ignore_threshold, drift_auto_threshold.
• Embedding-Based Semantic Search — Feature-level embeddings via fastembed + BGE-small-en-v1.5 (384 dimensions). search_node features mode now uses hybrid rank-based scoring (0.6 semantic + 0.4 lexical) with max-cosine similarity over per-feature vectors. Model auto-downloads on first search, runs fully offline afterward.
• TOML-driven paradigm pipeline — Framework detection and entity classification via declarative TOML configs instead of hardcoded patterns (#28)
• 7 additional language parsers — C, C++, Go (enhanced), Java (enhanced), with per-language entity and dependency modules (#28)
• Semantic drift re-routing and feature-based hierarchy routing (#30)
• Full-scale paper fidelity documentation
• MCP tool count: 15 → 17 (new: get_routing_candidates, submit_routing_decisions)

Changed

• Parser architecture refactored into per-language modules under crates/rpg-parser/src/languages/
• Paradigm detection moved to crates/rpg-parser/src/paradigms/
• graph_revision now uses updated_at timestamp instead of base_commit
• build_rpg, update_rpg, reload_rpg now clear pending routing and invalidate embeddings
• Test count: 275 → 379+

Fixed

• Windows build.rs path escaping and dropped Intel Mac from release matrix

[0.1.8] - 2026-02-09

Added

• Redux Toolkit frontend adapter for TypeScript/React/Next.js — extracts createSlice, createAsyncThunk, RTK Query hooks, and store configuration as first-class entities (#26)
• Benchmark comparison analysis (benchmarks/comparison.md)
• Use cases guide (use_cases.md)

[0.1.7] - 2026-02-08

Added

• .rpgignore support — exclude files from the RPG graph using gitignore-style patterns (#23)
• Rebuilt RPG graph with full semantic hierarchy and 100% lifting coverage

Changed

• Updated benchmark results for 652 entities, 39 queries

[0.1.6] - 2026-02-07

Added

• Enhanced TS/JS parser for React/Next.js patterns — JSX components, hooks, pages, layouts, API routes (#15)
• Per-file language detection in lifting for mixed-language projects (#21)
• Scoped dependency resolution during lifting (#21)

Fixed

• Entity IDs correctly rekeyed on file rename during incremental update (#19)
• New entities from renamed files receive hierarchy assignments (#19)
• npm publish made idempotent to avoid false-failure CI status (#14)

Changed

• Aligned MCP synthesis instructions with file synthesis prompt (#17)
• Fixed stale benchmark data (#17)

[0.1.5] - 2026-02-07

Added

• Paper alignment improvements — closer match to Algorithms 1-4 from the paper (#11)
• Deterministic edge ordering in serialized graph output (#11)
• Improved hierarchy construction prompts

[0.1.4] - 2026-02-06

Fixed

• Qualified entity IDs — use file:Class::method format to resolve merged-key coverage ceiling where entities from different files shared names (#8)
• 100% lifting coverage now achievable (557/557 entities)

[0.1.3] - 2026-02-06

Fixed

• Deterministic JSON serialization — all maps use BTreeMap for reproducible graph.json output across runs (#6)

Added

• Multi-repo setup instructions in README (#4)

[0.1.2] - 2026-02-06

Fixed

• Auto-preserve semantic features on build_rpg — previously, rebuilding the graph would discard all lifted features. Now, features from the old graph are automatically merged into the new graph for entities that still exist (#2)
• Collapsed nested if statements to satisfy clippy

Added

• npm publish in release workflow (CI)
• npm OIDC trusted publishers support

[0.1.1] - 2026-02-03

Initial npm release. Same code as v0.1.0, published to npm registry.

[0.1.0] - 2026-02-03

Initial public release. Independent Rust implementation of the RPG-Encoder framework described in arXiv:2602.02084.

Core Pipeline

• Semantic Lifting (Phase 1) — Parse code with tree-sitter, enrich entities with verb-object features via the connected coding agent's MCP interactive protocol (get_entities_for_lifting → submit_lift_results)
• Structure Reorganization (Phase 2) — Agent discovers functional domains and builds a 3-level semantic hierarchy (build_semantic_hierarchy → submit_hierarchy)
• Artifact Grounding (Phase 3) — Anchor hierarchy nodes to directories via LCA algorithm, resolve cross-file dependency edges (imports, invocations, inheritance)

Language Support

• 8 languages via tree-sitter: Python, Rust, TypeScript, JavaScript, Go, Java, C, C++
• Per-language entity extraction (functions, classes, methods, structs, traits, interfaces)
• Per-language dependency resolution (imports, calls, inheritance, trait impls)

Incremental Evolution

• Git-diff-based incremental updates (Algorithms 2-4 from the paper)
• Deletion pruning with hierarchy cleanup
• Modification with semantic drift detection (Jaccard distance)
• Structural entity insertion with dependency re-resolution
• Modified entities tracked for agent re-lifting

Navigation & Search

• search_node — Intent-based search across 3 modes: features, snippets, auto
• fetch_node — Entity details with source code, dependencies, hierarchy context; V_H hierarchy node fetch support
• explore_rpg — Dependency graph traversal (upstream/downstream/both) with configurable depth and edge filtering by kind (imports, invokes, inherits, contains)
• rpg_info — Graph statistics, hierarchy overview, per-area lifting coverage
• Cross-view traversal between V_L (code entities) and V_H (hierarchy nodes)
• TOON (Token-Oriented Object Notation) serialization for token-efficient LLM output

MCP Server

• 15 tools: build_rpg, search_node, fetch_node, explore_rpg, rpg_info, update_rpg, lifting_status, get_entities_for_lifting, submit_lift_results, finalize_lifting, get_files_for_synthesis, submit_file_syntheses, build_semantic_hierarchy, submit_hierarchy, reload_rpg
• Semantic lifting via connected coding agent — no API keys, no local LLMs, no setup
• Staleness detection on read-only tools (prepends [stale] notice when graph is behind HEAD)
• Auto-update on server startup when graph is stale (structural-only, sub-second)

CLI

• Commands: build, update, search, fetch, explore, info, diff, validate, export, hook
• --include / --exclude glob filtering for builds
• --since commit override for incremental updates
• Pre-commit hook: rpg-encoder hook install (auto-updates and stages graph on commit)
• Graph export as DOT (Graphviz) or Mermaid flowchart
• Graph integrity validation

Storage

• RPG graph committed to repos (.rpg/graph.json) — collaborators get instant semantic search without rebuilding
• Self-contained .rpg/.gitignore (ignores config.toml)
• Optional zstd compression for large graphs

Configuration

• .rpg/config.toml with sections: [encoding], [navigation], [storage]
• Environment variable overrides (RPG_BATCH_SIZE, RPG_SEARCH_LIMIT, etc.)
• Feature normalization: trim, lowercase, sort+dedup per paper spec

Code Quality

• Modular crate architecture: rpg-core, rpg-parser, rpg-encoder, rpg-nav, rpg-cli, rpg-mcp
• Clean cargo clippy --workspace --all-targets -- -D warnings
• Clean cargo fmt --all -- --check