Skip to content

CHANGELOG.md

Latest commit

 

History

History
122 lines (90 loc) · 8.48 KB

CHANGELOG.md

File metadata and controls

122 lines (90 loc) · 8.48 KB

Changelog

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

[Unreleased]

Features

  • --dry-run for generate — preview which files would be created or skipped without writing anything (agentrc generate mcp --dry-run)
  • JSONC support — agentrc.config.json, agentrc.eval.json, and policy files now accept // and /* */ comments
  • Array eval expectations — expectation field in eval configs accepts string[] for structured criteria

Bug Fixes

  • Config scaffolding for simple repos — agentrc init and the TUI now create agentrc.config.json even when no areas are detected, producing a minimal {} stub. Previously failed with "No areas detected. Cannot scaffold agentrc.config.json."

[2.1.0]

Features

  • Agent plugin & marketplace — expose built-in instruction-generation skills as a VS Code agent plugin (plugin/) with marketplace manifest (.github/plugin/marketplace.json). Skills can be installed from source or discovered through the marketplace.
  • SKILL.md-based prompts — replace hardcoded inline prompts with SKILL.md assets loaded via the Copilot SDK skillDirectories option and invoked with /skill-name prefixes.
  • Enhanced .NET detection — F# support, framework parsing, and expanded signals (#60)
  • De-branded terminology — --dry-run flag, batch instructions command (#55)

Bug Fixes

  • Fix instruction generation error: Dynamic require of "util" is not supported (#52)
  • Fix Windows .bat/.cmd shim spawn failure in copilotSdk.ts (#53)

Documentation

  • Rewrite README as concise landing page, add 8 user-facing doc pages (getting-started, concepts, commands, configuration, policies, ci-integration, at-scale, extension) (#62)
  • Add cross-platform guidance for process spawning and ESM/CJS interop

Dependencies

  • Bump simple-git production dependency (#49)
  • Bump github/gh-aw from 0.50.5 to 0.57.2 (#51)
  • Bump actions/download-artifact from 6.0.0 to 8.0.0 (#48)
  • Bump dev dependencies (#50)
  • Bump undici in vscode-extension (#59)

[2.0.0]

Complete Rewrite

AgentRC vNext is a complete rewrite as a TypeScript CLI tool (ESM, strict, ES2022) for priming repositories for AI-assisted development and evaluation.

New Commands

  • agentrc readiness — AI readiness report scoring repos across 9 pillars (style, build, testing, docs, dev-env, code-quality, observability, security, AI tooling) with a 5-level maturity model (Functional → Autonomous).
  • agentrc readiness --visual — GitHub-themed HTML report with light/dark toggle, expandable pillar details, and maturity model descriptions.
  • agentrc readiness --per-area — Per-area readiness scoring for monorepos with area-scoped criteria and aggregate thresholds.
  • agentrc readiness --policy — Customizable readiness policies (disable/override criteria, tune thresholds) via JSON, JS/TS, or npm packages; chainable with last-wins semantics.
  • agentrc batch-readiness — Consolidated visual readiness report across multiple repositories, with --policy support.
  • agentrc generate instructions — Generate copilot-instructions.md via Copilot SDK, with --per-app support for monorepos.
  • agentrc generate agents — Generate AGENTS.md guidance files.
  • agentrc instructions --areas — Generate file-based .instructions.md files scoped to detected areas with applyTo glob patterns.
  • agentrc eval --init — AI-powered eval scaffold generation that analyzes codebases and produces cross-cutting, area-aware eval cases.
  • agentrc eval --list-models — List available Copilot CLI models.
  • agentrc analyze — Standalone repo analysis command with structured --json output.

VS Code Extension

  • 8 command palette commands: Analyze, Generate Configs, Generate Instructions, AI Readiness Report, Run Eval, Scaffold Eval, Initialize Repository, Create PR.
  • Sidebar tree views: Analysis (languages, frameworks, monorepo areas) and Readiness (9-pillar scores with color-coded criteria).
  • Webview panels for readiness HTML reports and eval results.
  • Dynamic status bar showing detected languages after analysis.
  • PR creation with default-branch guard, selective file staging, and GitHub auth via VS Code API.
  • esbuild-bundled CJS output; CI typecheck and release-time VSIX packaging.

New Features

  • Azure DevOps integration — Full support for batch processing, PR creation, and repo cloning via Azure DevOps PAT authentication.
  • Headless automation — Global --json and --quiet flags on all commands; CommandResult<T> envelope with ok/status/data/errors. Headless batch mode via positional args or stdin piping.
  • Policy system — Layered policy chain for readiness reports: disable/override criteria, add extras, tune pass-rate thresholds. Config-sourced policies restricted to JSON-only for security.
  • Per-area readiness — 4 area-scoped criteria (area-readme, area-build-script, area-test-script, area-instructions) with 80% aggregate pass threshold.
  • File-based area instructions — .instructions.md files with YAML frontmatter (description, applyTo) for VS Code Copilot area scoping.
  • Expanded monorepo detection — Bazel (MODULE.bazel/WORKSPACE), Nx (project.json), Pants (pants.toml), Turborepo overlay, in addition to Cargo, Go, .NET, Gradle, Maven, npm/pnpm/yarn workspaces.
  • Smart area fallback — Large repos with 10+ top-level dirs automatically discover areas via heuristic scanning with symlink-safe directory traversal.
  • Eval trajectory viewer — Interactive HTML viewer comparing responses with/without instructions, including token usage, tool call metrics, and duration tracking.
  • Windows Copilot CLI support — .cmd/.bat wrapper handling via cmd /c, npm-loader.js detection, and CopilotCliConfig type replacing bare string paths.
  • Copilot CLI discovery — Cross-platform discovery with TTL caching and glob-based fallback for VS Code extension paths.
  • Centralized model defaults — Default model set to claude-sonnet-4.5 via src/config.ts.

Improvements

  • All file write paths now use safeWriteFile() — instructions, agents, and area files all reject symlinks and respect --force.
  • Unified agentrc pr command: both GitHub and Azure DevOps generate all three artifacts (instructions + MCP + VS Code configs) with consistent branch naming.
  • CommandResult<T> output envelope with structured JSON to stdout; human-readable output to stderr.
  • ProgressReporter interface for silent or human-readable progress across CLI and headless modes.
  • Symlink-safe directory scanning via isScannableDirectory() with lstat + realpath containment checks.
  • Path traversal protection via validateCachePath for cloned repo paths and double-layer defense for area applyTo patterns.
  • Credential sanitization in git push error messages to prevent token leaks.
  • buildAuthedUrl utility supporting both GitHub (x-access-token) and Azure DevOps (pat) auth.
  • checkRepoHasInstructions now re-throws non-404 errors instead of silently returning false.
  • init --yes now generates instructions, MCP, and VS Code configs (previously only instructions).
  • CSP meta tags added to eval and readiness HTML report generators.

Removed

  • Removed stub commands: templates, update, config.
  • Removed src/utils/cwd.ts — replaced by Copilot SDK workingDirectory session config.

Testing & Tooling

  • Vitest test framework with 267 tests across 13 test files covering analyzer, generator, git, readiness, visual report, fs utilities, cache path validation, policies, boundaries, CLI, output utilities, and PR helpers.
  • ESLint flat config with TypeScript, import ordering, and Prettier integration.
  • CI workflow with lint, typecheck, tests (Node 20/22, Ubuntu/macOS/Windows), build verification, and extension typecheck.
  • CI dogfooding: runs agentrc analyze --json and agentrc readiness --json on the repo itself.
  • Release automation via release-please with VSIX packaging for the VS Code extension.
  • Code coverage via @vitest/coverage-v8.

Project Setup

  • Added CONTRIBUTING.md, SECURITY.md, LICENSE (MIT), and CODEOWNERS.
  • Added issue templates (bug report, feature request) and PR template.
  • Added .github/agents/ with multi-model code review agents (Opus, Gemini, Codex).
  • Added .github/prompts/ with reusable prompts (deslop, review, generate-improvements).
  • Added examples folder with sample eval config and CLI usage guide.
  • Added .prettierrc.json with project formatting rules.