Skip to content

Changelog

All notable changes to apexe are documented in this file.

Format follows Keep a Changelog. Versioning follows Semantic Versioning.

[0.2.0] - 2026-03-29

Added

  • 6 new apexe serve flags — --tags, --prefix, --acl, --enable-approval, --no-logging, --skip-validation for full McpServerBuilder control from CLI.
  • Expanded help fallback — Scanner tries --help all and -h when --help yields few flags. Fixes curl (0 → 12 flags on macOS).
  • GNU regex relaxation — Flag regex now matches 1-space indent (\s{1,}) for curl-style help format.
  • apexe --man — Generates complete roff man page via apcore_cli::build_program_man_page(), including commands, flags, exit codes, and docs URL.
  • Documentation URL — set_docs_url() sets https://github.com/aiperceivable/apexe in help/man output.
  • Explorer allow_execute — serve_with_options() enables tool execution from Explorer UI.
  • Env var test stability — Global ENV_LOCK Mutex prevents parallel test race conditions on environment variables.

Changed

  • Dependencies switched to crates.io — apcore = "0.14", apcore-cli = "0.4", apcore-mcp = "0.11", apcore-toolkit = "0.4". Zero path dependencies.
  • ACL opt-in only — --acl <path> required to enable access control. Without it, all tools are allowed (fixes Explorer AclDenied issue).
  • require_auth removed — Was silently ineffective without JWT authenticator. Removed to prevent false security assumptions.
  • Config override validation — CLI scan_depth (1-5) and timeout (>0) overrides now range-checked with warning on rejection.

Fixed

  • Explorer UI empty response — Fixed by using serve_with_options() with ExplorerOptions { explorer: true, allow_execute: true } instead of serve().
  • clippy result_large_err — Suppressed on execute_subprocess (Rust 1.94 stricter closure checking).
  • Flaky env var tests — Eliminated race condition with ENV_LOCK mutex.

[0.1.0] - 2026-03-28

First release with full apcore ecosystem integration.

Added

Scanning Engine - Three-tier deterministic CLI scanner: --help parsing (GNU, Click, Cobra, Clap), man page enrichment (DESCRIPTION + OPTIONS sections), shell completion subcommand discovery - ParserPipeline with automatic format detection and priority routing - SubcommandDiscovery with recursive scanning up to depth 5 - ScanCache with JSON filesystem caching and version-based invalidation - ToolResolver with binary path resolution and version detection - Plugin system via CliParser trait for custom parser registration - Tier 3 completion-discovered subcommands merged back into scan results

Adapter Layer (ScannedCLITool → ScannedModule) - CliToolConverter: recursive subcommand tree flattening, dot-separated module IDs (cli.git.commit) - JSON Schema generation with full type mapping: String, Integer, Float, Boolean, Path (format: path), URL (format: uri), Enum (enum: [...]) - Repeatable flags → array schemas, required flags → required array, default value coercion - Behavioral annotation inference from command names (readonly/destructive patterns) and flag boosting (--force → requires_approval, --dry-run → idempotent) - DisplayResolver integration: auto-generated MCP aliases (cli.git.commit → git_commit), per-surface display metadata

Module Executor - CliModule: implements apcore Module trait for CLI subprocess execution - Async execution via tokio::task::spawn_blocking with tokio::time::timeout - Shell injection prevention: 15-character blacklist validated at construction time and runtime - Context integration: trace_id, identity, duration_ms propagated through execution - ai_guidance on non-zero exit codes with stderr context for AI self-correction

MCP Server - McpServerBuilder: fluent API for module loading → Registry → Executor → APCoreMCP - Transports: stdio (Claude Desktop/Cursor), streamable-http, SSE - LoggingMiddleware enabled by default (structured logging with redaction) - ElicitationApprovalHandler for interactive destructive command approval (opt-in) - Tags and prefix filtering for tool access control - Explorer UI support (HTTP transports) - export_openai_tools(): OpenAI function calling format export - Config snippet generation for Claude Desktop and Cursor

Governance - AclManager: wraps apcore::ACL, auto-generates default-deny rules from annotations - AuditManager: wraps apcore_cli::AuditLogger, JSONL append-only with SHA-256 input hashing - SandboxManager: wraps apcore_cli::Sandbox, subprocess isolation with timeout enforcement

Output - YamlOutput: wraps apcore_toolkit::YAMLWriter with optional verification - load_modules_from_dir(): reads .binding.yaml files as Vec<ScannedModule>

Configuration - 4-tier config resolution: CLI flags > env vars > config file > defaults - 5 environment variables: APEXE_MODULES_DIR, APEXE_CACHE_DIR, APEXE_LOG_LEVEL, APEXE_TIMEOUT, APEXE_SCAN_DEPTH - Optional apcore::Config integration via ~/.apexe/apcore.yaml - Range validation on all numeric config overrides

Error Handling - From<ApexeError> for ModuleError: all 9 error variants mapped with ai_guidance - into_module_error_with_trace() for trace_id attachment - Rich error display in CLI with suggestion text

CLI - apexe scan <TOOLS>... — scan with --depth, --no-cache, --format, --output-dir - apexe serve — --transport, --host, --port, --explorer, --name, --show-config - apexe list — --format, --modules-dir - apexe config — --show, --init

Documentation - Quick Start guide (docs/quickstart.md) - Full User Manual with 15 chapters (docs/user-manual.md) - Technical Design document (docs/apcore-integration/tech-design.md) - 7 feature specifications (docs/features/v2-f1..f7) - Feature Manifest with module map (docs/FEATURE_MANIFEST.md)

Examples - examples/basic/ — Shell script walkthrough: scan → list → serve - examples/programmatic.rs — Rust library API: scan → convert → export → build server

Dependencies

Crate Version Role
apcore 0.14 Core types: Module, Registry, ACL, ModuleError, Config
apcore-toolkit 0.4 ScannedModule, YAMLWriter, DisplayResolver, Verifier
apcore-mcp 0.11 MCP server: APCoreMCP, transports, auth, Explorer
apcore-cli 0.3 AuditLogger, Sandbox

Stats

  • 39 source files, ~8,850 lines of Rust
  • 338 tests, 0 failures
  • All quality gates pass: cargo fmt, cargo clippy -D warnings, cargo test --all-features