v4.8.1 - Resilient JSON Output Extraction & Headless Claude Permissions

Release v4.8.1

Fixed

• Resilient JSON Output Extraction (_extract_json):
  • Fixed an issue where reviewer verdicts and triage responses wrapped in conversational preambles (e.g. Confirmed: HEAD equals base commit... {"verdict": ...}), markdown code fences, or trailing commentary caused strict json.loads parsing failures (Expected a JSON object with verdict...).
  • Added robust 4-stage JSON extraction: fast-path direct parsing, markdown code fence extraction, outermost brace pair detection, and balanced brace depth scanner.
• Claude Headless Batch Mode Permissions (�cceptEdits):
  • Fixed Claude provider invocation in modifying stages (orchestrator / integrator). Changed --permission-mode default (which prompts interactively on stdin and blocks headless execution) to --permission-mode acceptEdits, enabling Claude to write deliverables and execute repository modifications without hanging.
  • Read-only stages (reviewer / verifier / triage) continue to strictly enforce --permission-mode plan.
  • Propagated --dangerously-skip-permissions to Claude when ullAuto: true is configured.
• Subprocess stdin Disconnection (stdin=DEVNULL):
  • Redirected spawned agent subprocess stdin to subprocess.DEVNULL, eliminating Claude's 3-second stdin timeout warning (
    o stdin data received in 3s) and avoiding hangs in headless CI/CD and terminal execution.

v4.8.0 - 360° Application Audit Engine

Added

• Native \�i-team audit [project]\ CLI command with flags: --output, --lang, --worktree, --solo, and --auto-merge\
• \�udit/full-app-audit\ skill for Antigravity (.agents) and Claude Code (.claude) in both Polish and English
• Dedicated \�uditor\ persona role in .agents/agents/auditor.md\ and .claude/agents/auditor.md\
• Automatic inclusion of the audit skill in all project profiles (\core, \python, \web, \postgres, \ocr, \geneteka, \ull)
• Setup script updates (\setup-project.sh\ and \setup-project.ps1) with Option C for instant 360° audits

v4.7.0: One-Command Project Onboarding & Setup Script

Added

• One-Command Project Onboarding (\�i-team onboard\ / \�i-team quickstart\ / \�i-team setup):
  • Automatically initializes Git (if missing), installs or updates \�i-team\ with stack detection.
  • Automatically resolves template conflicts with --strategy upstream.
  • Automatically configures \�i-team.config.json\ with safe defaults (\�llowUnreviewedLowRisk: true, smart test verification for npm/pytest/cargo/go, solo/multi provider rules).
  • Automatically cleans \PROJECT_CONTEXT.md\ (removes \TODO\ placeholders with inferred metadata).
  • Automatically commits changes to Git and runs full healthcheck diagnostics (\�i-team doctor).
• Automated Onboarding Scripts (\setup-project.sh\ & \setup-project.ps1):
  • Universal Bash and PowerShell scripts enabling instant zero-config setup across multiple repositories with a single command.
• Bulk Conflict Resolution (\�i-team resolve all --strategy <keep|upstream>):
  • Resolves all pending template update conflicts in a single command.

v4.6.0: Single-Provider Solo Mode and Reviewer Availability Fallback

Release v4.6.0

Highlights

• Single-Provider Solo Mode (--solo / --single-provider / "singleProvider": true):
  • Run the entire AI Engineering Team process on a single provider (agy / Gemini 3.8 Flash High) without requiring external CLI tools (codex, claude) or external quotas.
  • Review policy dynamically adapts: in solo mode, independent review roles execute using isolated read-only primaryProvider processes (--mode plan, --sandbox, zero write permissions).
  • Added --solo flag to ai-team doctor to verify system readiness specifically for solo workflows without flagging missing multi-provider CLIs.
• Reviewer Availability Fallback (availabilityFallback):
  • Automatically handles external reviewer CLI failures (quota exhausted, HTTP 429 rate limit, authentication issues, or network timeouts).
  • When an external reviewer fails at runtime, the runner catches the error, logs [FALLBACK], and seamlessly executes the review using an isolated read-only primary provider (agy).
  • Records fallbackFrom in review reports and produces clean isolated stage artifacts (review-<round>-<reviewer>-fallback.json).
  • Configurable via "availabilityFallback": true/false in ai-team.config.json or CLI flags --availability-fallback / --no-availability-fallback.
• Worktree Lifecycle Fix:
  • Eliminated premature return in isolated worktree runs, ensuring automatic merge and interactive deployment prompts execute reliably.

Verification

• 271 pytest unit and integration tests passing (100% pass rate).
• Full bilingual English/Polish documentation and schema consistency.

AI Engineering Team v4.5.0

Release v4.5.0 - Automated Updates Workflow, Repository Git Hygiene & Upstream Version Check

Added

• Automated Upstream Updates Workflow (ai-team workflow [project]):
  • Installs a ready-to-use GitHub Actions workflow (.github/workflows/ai-team-update.yml) in target projects.
  • Automatically runs on a weekly schedule (Mondays 06:00 UTC) or on manual trigger (workflow_dispatch).
  • Upgrades ai-engineering-team, runs ai-team update ., and creates a clean Pull Request (chore/ai-team-update) when template or skill updates are available.
• Enhanced Repository Git Hygiene (ai-team gitignore [project] [--private]):
  • Automatically provisions .ai-team/.gitignore during installation and updates to protect temporary merge conflicts and backups from repository pollution.
  • ai-team gitignore: Configures standard runtime ignore rules (.ai/runs/, .ai-team/conflicts/, etc.) in the project's root .gitignore.
  • ai-team gitignore --private: Configures a completely private solo developer mode in .git/info/exclude so all AI files remain strictly local and invisible to remote team repositories.
• Upstream Version Checking (ai-team update --check and ai-team status):
  • Offline-safe check that queries GitHub Releases to notify developers when newer versions are available, with actionable upgrade instructions.

AI Engineering Team v4.4.0

Release v4.4.0 - Proactive Autonomous Skill Provisioning by AI Agents

Added

• Proactive & Autonomous Skill Provisioning by AI Agents:
  • The AI Orchestrator now automatically and autonomously manages skills without requiring user intervention or manual CLI commands.
  • Pre-Implementation Stage: Prior to starting work, the runner inspects project tech stack markers and task prompt intent to automatically install missing matching skills from the catalog.
  • Post-Implementation Stage: Inspects files touched or introduced during implementation diff (e.g. Dockerfiles, SQL schemas, migrations) and automatically provisions matching skills before independent review rounds.
  • Autonomous Project Skills: Instructed agents in AI_TEAM.md and AGENTS.md to formulate and maintain project-specific conventions and patterns under .agents/skills/project/<name>/SKILL.md.
  • Configuration & Opt-Out: Enabled by default ("autoSkills": true), with optional manual opt-out via --no-auto-skills flag in ai-team run or setting "autoSkills": false in ai-team.config.json.

AI Engineering Team v4.3.0

Release v4.3.0: Intelligent tech stack detection and auto-install (ai-team install --auto), dedicated skill management CLI (ai-team skill list/suggest/add/remove), expanded catalog (devops/docker-quality, security/secure-coding), and task-based dynamic context routing.

AI Engineering Team v4.2.0

Release v4.2.0: Automated merge prompt after worktree isolation, auto-merge flags, and clean branch disposal to prevent repository clutter.

AI Engineering Team v4.1.0

Release v4.1.0: Git worktree isolation (--worktree), interactive review CLI (ai-team review), AST codebase mapping (ai_team.repomap), and dynamic learnings memory (.ai/LEARNINGS.md).

AI Engineering Team v4.0.0

Release v4.0.0: Framework hardening, bilingual support, process-tree isolation, and audit regression test suite.