prjct

Context layer for AI coding agents.

Works with Claude Code, Gemini CLI, Antigravity, Cursor IDE, and more.

What is prjct?

prjct gives AI coding agents the context they need about your project. It maintains state between sessions, tracks progress, and ensures agents understand your codebase.

Your AI Agent                              prjct
(Claude/Gemini/Antigravity/Cursor)           │
         │                                   │
         │  "What am I working on?"          │
         │ ────────────────────────────────► │
         │                                   │ Reads project context
         │  Task: "Add user auth"            │
         │  Branch: feature/auth             │
         │  Subtask 2/5: API routes          │
         │ ◄──────────────────────────────── │
         │                                   │
         ▼                                   │
    Writes code with full context            │

Install

npm install -g prjct-cli

Quick Start

Claude Code / Gemini CLI

# 1. One-time global setup
prjct start

# 2. Initialize your project
cd my-project
prjct init

# 3. Open in Claude Code or Gemini CLI and use:
p. sync                    # Analyze project
p. task "add user auth"    # Start a task
p. done                    # Complete subtask
p. ship                    # Ship with PR

Google Antigravity

# 1. One-time global setup (installs prjct as a skill)
prjct start

# 2. Initialize your project
cd my-project
prjct init

# 3. Open in Antigravity and use:
p. sync                    # Analyze project
p. task "add user auth"    # Start a task
p. done                    # Complete subtask
p. ship                    # Ship with PR

Note: prjct integrates as a Skill (not MCP server) for zero-overhead operation.

Cursor IDE

# 1. Initialize your project (no global setup needed)
cd my-project
prjct init

# 2. Open in Cursor and use:
/sync                      # Analyze project
/task "add user auth"      # Start a task
/done                      # Complete subtask
/ship                      # Ship with PR

Note: Cursor uses /command syntax. Commands are installed per-project in .cursor/commands/. If deleted, run /sync to regenerate.

Core Workflow

Claude/Gemini/Antigravity:  p. sync  →  p. task "..."  →  [code]  →  p. done  →  p. ship
Cursor:                     /sync    →  /task "..."    →  [code]  →  /done    →  /ship

How It Works

| Component | Claude Code | Gemini CLI | Antigravity | Cursor IDE | |-----------|-------------|------------|-------------|------------| | Router | ~/.claude/commands/p.md | ~/.gemini/commands/p.toml | Skill | .cursor/commands/*.md | | Config | ~/.claude/CLAUDE.md | ~/.gemini/GEMINI.md | ~/.gemini/antigravity/skills/prjct/ | .cursor/rules/prjct.mdc | | Storage | ~/.prjct-cli/projects/ | ~/.prjct-cli/projects/ | ~/.prjct-cli/projects/ | ~/.prjct-cli/projects/ | | Scope | Global | Global | Global | Per-project | | Syntax | p. command | p. command | p. command | /command |

All agents share the same project storage, so you can switch between them freely.

Commands

| Claude/Gemini | Cursor | Description | |---------------|--------|-------------| | p. sync | /sync | Analyze project, generate domain agents | | p. task "desc" | /task "desc" | Start task with auto-classification | | p. done | /done | Complete current subtask | | p. ship "name" | /ship "name" | Ship feature with PR + version bump | | p. pause | /pause | Pause current task | | p. resume | /resume | Resume paused task | | p. bug "desc" | /bug "desc" | Report a bug | | p. linear | - | Linear integration | | p. github | - | GitHub Issues integration |

Task History

prjct automatically tracks your completed tasks to help AI agents learn from patterns and make better decisions across sessions.

How It Works

When you complete a task (p. done / /done), prjct stores:

• Task description and classification (feature, bug, improvement, chore)
• Start and completion timestamps
• Number of subtasks and their summaries
• Git branch name and Linear issue ID (if linked)
• PR URL (if shipped)

This history is:

• Bounded: Maximum 20 entries with FIFO (First-In-First-Out) eviction
• Contextual: Filtered by task type when starting similar work
• Persistent: Survives across sessions and agent types

Context Injection

Task history is automatically injected into the AI agent's context:

• When starting a task: Shows 3 most recent tasks of the same type (e.g., recent bug fixes when starting a new bug)
• When idle: Shows 5 most recent tasks across all types
• Purpose: Helps agents identify patterns, avoid repeating mistakes, and build on previous solutions

Accessor Methods (for developers)

import { stateStorage } from './storage/state-storage'

// Get full task history (max 20 entries, newest first)
const history = await stateStorage.getTaskHistory(projectId)

// Get most recent completed task
const recent = await stateStorage.getMostRecentTask(projectId)

// Get tasks by classification
const bugs = await stateStorage.getTaskHistoryByType(projectId, 'bug')
const features = await stateStorage.getTaskHistoryByType(projectId, 'feature')

CLI Commands

prjct start          # First-time setup (Claude/Gemini)
prjct init           # Initialize project (+ Cursor setup)
prjct sync           # Analyze project and generate context
prjct verify         # Verify analysis integrity (cryptographic)
prjct verify --semantic  # Verify analysis consistency (semantic)
prjct --version      # Show version + provider status
prjct --help         # Show help

Analysis Verification

prjct provides two types of analysis verification to ensure data integrity and logical consistency:

Cryptographic Verification (Default)

prjct verify [--json]

Verifies the integrity of sealed analysis results using cryptographic signatures. This ensures:

• Analysis data hasn't been tampered with
• Sealed analysis matches the original analysis
• Hash signatures are valid

When to use: After sealing an analysis (prjct seal) to confirm data integrity.

Semantic Verification (PRJ-270)

prjct verify --semantic [--json]

Validates that analysis results match the actual project state. This checks:

• ✓ Frameworks exist in package.json dependencies
• ✓ Languages match actual file extensions (.ts → TypeScript)
• ✓ Pattern locations reference real files in the project
• ✓ File count is accurate (within 10% tolerance)
• ✓ Anti-pattern files exist when referenced

When to use: Before sealing an analysis to catch logical inconsistencies or after project changes to validate analysis accuracy.

Example output:

Semantic Verification Report
────────────────────────────
  ✓ Framework verification: passed (2 frameworks validated)
  ✓ Language verification: passed (1 language validated)
  ✓ Pattern locations: passed (12 patterns verified)
  ✓ File count verification: passed (324 files, within tolerance)
  ✓ Anti-pattern files: passed (3 anti-patterns verified)

Result: PASSED (5/5 checks)
Total time: 145ms

Both verification modes support --json flag for programmatic use.

Environment Variables

Configuration

| Variable | Default | Description | |----------|---------|-------------| | PRJCT_CLI_HOME | ~/.prjct-cli | Override global storage location. Useful for tests or sandboxed environments. | | PRJCT_DEBUG | (unset) | Enable debug logging. Values: 1, true, or a log level (error, warn, info, debug). | | DEBUG | (unset) | Fallback debug flag (used if PRJCT_DEBUG is not set). Values: 1, true, or prjct. | | CI | (unset) | Set automatically in CI environments. Skips interactive prompts. |

JIRA Integration

| Variable | Default | Description | |----------|---------|-------------| | JIRA_BASE_URL | (none) | JIRA instance URL (e.g., https://myorg.atlassian.net). | | JIRA_EMAIL | (none) | Email for JIRA API authentication. | | JIRA_API_TOKEN | (none) | API token for JIRA authentication. Generate at Atlassian API tokens. |

Agent Detection (Auto-set)

These are typically set by the AI agent runtime, not by users:

| Variable | Description | |----------|-------------| | CLAUDE_AGENT | Set when running inside Claude Code. | | ANTHROPIC_CLAUDE | Alternative Claude environment indicator. | | MCP_AVAILABLE | Set when MCP (Model Context Protocol) is available. | | HOME / USERPROFILE | Standard OS home directory (used for path resolution). |

Usage Examples

# Enable debug logging
PRJCT_DEBUG=1 prjct sync

# Use a custom storage location
PRJCT_CLI_HOME=/tmp/prjct-test prjct init

# Configure JIRA integration via env vars
export JIRA_BASE_URL=https://myorg.atlassian.net
export [email protected]
export JIRA_API_TOKEN=your-api-token
prjct jira setup

Requirements

• Node.js 18+ or Bun 1.0+
• One of: Claude Code, Gemini CLI, Antigravity, or Cursor IDE

Links

• Website
• GitHub
• npm

License

MIT