Antigravity agent harness

buddhi init can scaffold an Antigravity harness in .agents/. The harness connects a workspace to buddhi-mcp and provides reusable workflows, specialist agent definitions, rules, hooks, templates, and a memory index. Existing customized harness files are preserved when buddhi init runs again.

What gets scaffolded

Path	Purpose
`.agents/mcp_config.json`	Registers the workspace MCP server.
`.agents/skills/`	Provides slash-invoked workflows and supporting skills.
`.agents/agents/`	Contains specialist agent definitions.
`.agents/rules/`	Stores project-level agent rules.
`.agents/hooks.json` and `.agents/hooks/`	Configure lifecycle hooks, including the destructive-command guard.
`.agents/templates/`	Holds specification, plan, task, review, and agent templates.
`.agents/memory/`	Contains the memory index and topic files.

The exact set of bundled files can change between BuddhiAI CLI releases. Inspect the generated .agents/ directory in the target project.

Spec-Driven Development

Use the SDD workflow to describe and verify a change before considering it complete:

/specify creates a feature specification with prioritized, testable user stories.
/spec-plan grounds an architectural plan in the repository and consults domain specialists.
/spec-tasks turns the spec and plan into phased implementation tasks.
/implement executes tasks and records progress.
/verify runs the project checks and maps results to acceptance criteria.

buddhi sdd provides supporting helpers used by these workflows. See the command reference.

Other bundled workflows

/ask refreshes and queries the code graph to explain project structure.
/agents-md creates or updates project instructions while preserving custom sections.
/quick-plan prepares a lightweight implementation plan.
/document-codebase generates or refreshes OKF symbol documentation.
/debug investigates a reported issue and prepares a root-cause explanation.
/remember records durable project conventions, decisions, or preferences.
/status reports the harness state, including documentation freshness and open plans.

The CLI also bundles skills for design systems, accessibility, anti-slop review, testing, refactoring, documentation, performance, and visual review. Examples include /design-md, /a11y, /antislop, /write-tests, /refactor-clean, /write-docs, /audit-performance, and /compare-screenshots. See the .agents/skills/ directory created in the project for the complete installed list and each skill’s instructions.

Specialist agents

The harness includes read-only specialists for frontend, backend, database, testing, security, deployment, and Git work. The /spec-plan and /quick-plan workflows can dispatch relevant specialists; terminal-runner is used for shell commands, builds, and tests.

Rules, hooks, and memory

Rules provide shared guidance for graph and OKF documentation use, safety, accessibility, and code conventions.
The PreToolUse hook in .agents/hooks/guard_destructive.py blocks selected destructive operations, including force pushes, hard resets, and database drop or truncate commands. Treat it as a safety backstop, not a replacement for reviewing a command.
.agents/memory/MEMORY.md indexes project memory topics. The buddhi memory commands manage structured concepts, source references, and index updates; see Project memory.

A typical workflow

Install BuddhiAI CLI and initialize a repository with buddhi init.
Open that repository in Google Antigravity and confirm the workspace MCP server is available.
Run /specify for a change that needs an explicit specification.
Use /spec-plan, /spec-tasks, and /implement to plan and execute the work.
Run /verify and review the reported evidence.
Refresh the graph with buddhi generate when project code changes.