AI代码智能运行时

The LLM should not own repository context.
CodeAware should.

Do not make the LLM rediscover the repository.
Compile the repository into reusable semantic intelligence first.

Repository
→ Discovery
→ AST Parsing
→ Semantic Extraction
→ SemanticIndex
→ SemanticContextAssembler
→ ContextPackage
→ Agent
→ Trace
→ Recovery
→ Architecture Memory
→ Semantic Routing

src/v4/
  architecture_memory.rs
  budget.rs
  cache.rs
  cache_invalidation.rs
  call_graph.rs
  context.rs
  context_items.rs
  contracts.rs
  discovery.rs
  errors.rs
  impact.rs
  import_graph.rs
  index_builder.rs
  language_support.rs
  precision.rs
  ranking.rs
  recovery.rs
  retrieval.rs
  semantic_context.rs
  semantic_index.rs
  semantic_router.rs
  semantic_tools.rs
  storage.rs
  summaries.rs
  symbols.rs
  tests_graph.rs
  tokens.rs
  tools.rs
  trace.rs

codeaware.get_task_context
codeaware.find_symbol
codeaware.find_callers
codeaware.find_tests
codeaware.diff_impact

{
  "jsonrpc": "2.0",
  "id": 10,
  "method": "tools/call",
  "params": {
    "name": "codeaware.get_task_context",
    "arguments": {
      "repo_root": "/workspace/project",
      "goal": "Refactor semantic context assembly"
    }
  }
}

{
  "jsonrpc": "2.0",
  "id": 11,
  "method": "tools/call",
  "params": {
    "name": "codeaware.find_symbol",
    "arguments": {
      "repo_root": "/workspace/project",
      "query": "ContextPackage"
    }
  }
}

{
  "jsonrpc": "2.0",
  "id": 12,
  "method": "tools/call",
  "params": {
    "name": "codeaware.find_callers",
    "arguments": {
      "repo_root": "/workspace/project",
      "symbol": "build_context"
    }
  }
}

{
  "jsonrpc": "2.0",
  "id": 13,
  "method": "tools/call",
  "params": {
    "name": "codeaware.find_tests",
    "arguments": {
      "repo_root": "/workspace/project",
      "symbol": "ContextPackage"
    }
  }
}

{
  "jsonrpc": "2.0",
  "id": 14,
  "method": "tools/call",
  "params": {
    "name": "codeaware.diff_impact",
    "arguments": {
      "repo_root": "/workspace/project",
      "changed_path": "src/v4/tools.rs"
    }
  }
}

Cursor / Claude Code / Gemini CLI / OpenCode
        ↓
CodeAware MCP
        ↓
SemanticIndex + ContextPackage + Budget + Trace
        ↓
Repository

codeaware.get_task_context(goal="Fix login session handling")

codeaware.find_symbol(query="ContextPackage")

codeaware.find_callers(symbol="build_context")

codeaware.find_tests(symbol="ContextPackage")

codeaware.diff_impact(changed_path="src/v4/tools.rs")

Agent requests context.
CodeAware decides what context is allowed.

symbols → imports → calls → tests → impact

max files read
max files changed
max tool calls
max context tokens
stop conditions

Why was this file selected?
Why was this path excluded?
How many estimated tokens were used?

Implemented: yes
Documented: yes
MCP-dispatch wired: yes
CI workflow added: yes
CI green: must be verified from GitHub Actions after workflow execution

cargo fmt --all -- --check
cargo clippy --all-targets --all-features -- -D warnings
cargo test --all --all-features
cargo build --release --all-features

git clone https://github.com/mhmtbsbyndr/codeaware-mcp.git
cd codeaware-mcp

cargo build --release

./target/release/codeaware-mcp

cargo test

./target/release/codeaware-mcp

chmod +x scripts/setup-codeaware-mcp-dashboard-launchd.sh
./scripts/setup-codeaware-mcp-dashboard-launchd.sh install /usr/local/bin/codeaware-mcp

./scripts/setup-codeaware-mcp-dashboard-launchd.sh install /home/$USER/.cargo/bin/codeaware-mcp

{
  "mcpServers": {
    "codeaware": {
      "command": "/absolute/path/to/codeaware-mcp/target/release/codeaware-mcp",
      "args": [],
      "env": {}
    }
  }
}

rustup update
cargo build --release
cargo test

Read("src/server.rs")       -> hundreds of source lines
Run("cargo test")           -> hundreds of noisy log lines
Read("src/server.rs") again -> same content again
Context compaction           -> working memory disappears

codeaware.get_task_context   -> bounded semantic context package
codeaware.find_symbol        -> symbol-level retrieval
codeaware.find_callers       -> caller graph lookup
codeaware.find_tests         -> related tests
codeaware.diff_impact        -> impact-aware reasoning
semantic_router              -> cheap/balanced/premium model routing hint
semantic_recovery            -> compact task recovery snapshot

AI Coding Agent
      |
      v
MCP JSON-RPC / stdio
      |
      v
codeaware-mcp Runtime
      |
      +-- Stable Compression Layer
      |     +-- smart_read
      |     +-- smart_run
      |     +-- git intelligence
      |
      +-- v4 Persistent Code Intelligence Kernel
      |     +-- task contracts
      |     +-- budget engine
      |     +-- discovery/ranking
      |     +-- summaries/token estimation
      |     +-- context packages
      |     +-- semantic index
      |     +-- symbols/imports/calls/tests
      |     +-- impact analysis
      |     +-- architecture memory
      |     +-- semantic recovery
      |     +-- semantic router
      |
      +-- Token Runtime
      |     +-- token_stats
      |     +-- token_savings_report
      |     +-- benchmark_compression
      |
      +-- Context Optimization Runtime
      |     +-- get_relevant_code
      |     +-- code_search
      |     +-- get_relevant_test_errors
      |     +-- get_project_context
      |     +-- tool_manager
      |
      +-- Progressive Memory Foundation
      |     +-- compact memory index
      |     +-- timeline window
      |     +-- observation details
      |     +-- privacy tag filtering
      |     +-- memory citations
      |
      +-- Safety Foundation
            +-- security policy
            +-- command validation
            +-- path validation
            +-- MCP routing

cargo test

./target/release/codeaware-mcp

{"jsonrpc":"2.0","id":1,"method":"initialize","params":{}}

{"jsonrpc":"2.0","id":2,"method":"tools/call","params":{"name":"token_stats","arguments":{}}}

{"jsonrpc":"2.0","id":3,"method":"tools/call","params":{"name":"codeaware.get_task_context","arguments":{"repo_root":".","goal":"Explain v4 semantic context"}}}

docs/CODEAWARE_V4_MASTERPLAN.md
docs/CODEAWARE_V4_ROADMAP.md
docs/CODEAWARE_V4_PHASE1_IMPLEMENTATION_SPEC.md
docs/CODEAWARE_V4_IMPLEMENTATION_SNAPSHOT.md
docs/CODEAWARE_V4_PHASE2_STATUS.md
docs/CODEAWARE_V4_FINAL_ARCHITECTURE.md
docs/CODEAWARE_V4_MCP_TOOLS.md

- run full cargo test in CI/local environment
- fix any compile/test regressions if found
- extend tree-sitter support beyond Rust extraction
- add production-grade AST call extraction
- improve semantic index cache invalidation strategy