Crosslink

You direct. Your agents remember. Nothing gets lost.

AI coding agents forget everything between conversations. Crosslink gives them persistent memory — sessions, handoff notes, issue tracking, and breadcrumbs that survive context compression, session restarts, and agent handoffs.

Why Crosslink?

You tell your agent to fix a bug. It investigates, finds the root cause, starts a fix — then the context window fills up. The agent restarts with zero memory. You explain everything again.

Crosslink eliminates this. Your agents track their own work, leave handoff notes for the next session, and coordinate with other agents on the same repo. You stay in the driver's seat. They handle the bookkeeping.

Quick Start

You say:

"Fix the auth token refresh bug — high priority"

Your agent handles the rest:

crosslink quick "Fix auth token refresh" -p high -l bug
crosslink session action "Exploring auth module..."
# ... investigates, implements, tests, commits ...
crosslink session end --notes "Fixed token refresh. PR ready for review."

Next session, any agent picks up right where this one left off.

Manual CLI equivalent

cargo install crosslink
crosslink init
crosslink session start
crosslink quick "Fix auth token refresh" -p high -l bug
# ... do the work yourself ...
crosslink session end --notes "Fixed token refresh."

Features

Core Issue Tracking

• Session memory — Handoff notes, breadcrumbs, and session state survive restarts and context compression
• Local-first — All data in SQLite (.crosslink/issues.db), no cloud, works offline
• Smart workflow — quick creates and starts work in one step. next recommends what to tackle. tree shows the hierarchy
• Subissues & dependencies — Break tasks down, track blocking relationships
• Time tracking, milestones, archiving — Full project management in the CLI
• Templates — Built-in templates for bugs, features, refactors, and research

Design Document Workflow

Turn ideas into validated, codebase-grounded specs before writing code.

• /design — Interactive design document authoring through explore → interview → draft → validate loop
• Codebase-grounded — Explores real code, asks questions informed by what it finds, references real files
• Validation — Requirements, acceptance criteria, and open questions tracked in the document
• Iteration — /design --continue <slug> to refine an existing design across sessions
• Feeds into implementation — Design docs drive kickoff --doc and swarm init --doc

Multi-Agent Orchestration

Launch multiple agents and let them coordinate automatically.

• Distributed locking — Agents claim issues via a shared git coordination branch, preventing conflicts
• Agent identity — Each agent gets a unique ID and SSH signing key (crosslink agent init)
• crosslink kickoff — Launch autonomous agents in isolated git worktrees
  • Agents explore, implement, test, commit, and self-review — fully tracked through crosslink
  • --verify local|ci|thorough — Configurable verification levels from local tests to CI + adversarial self-review
  • Design doc-driven: pass --doc to generate implementation plans from a design document
  • kickoff plan — Read-only gap analysis against the codebase before committing to a build
  • kickoff report — Spec validation reports from completed agents
  • kickoff graph — Visual branch topology of active kickoff branches
  • kickoff list / kickoff cleanup — Monitor and manage running agents
• crosslink swarm — Multi-agent phased builds from design documents
  • swarm init --doc — Decompose a design document into phases and work units
  • swarm launch / swarm gate / swarm checkpoint — Execute, gate, and record phase progress
  • swarm resume — Reconstruct state and continue after interruption
  • Budget-aware scheduling with cost estimation and configurable window duration
  • Multi-swarm support — manage multiple independent swarms with swarm create, swarm list, swarm switch
  • Plan editing — restructure plans post-init with move, merge, split, remove, reorder, rename
• Container execution — Run agents in isolated Docker containers (crosslink container)

Knowledge Management

Research done by one agent is available to all.

• crosslink knowledge — CRUD for markdown knowledge pages with YAML frontmatter
• Full-text search — knowledge search across all pages with tag and date filtering
• Cross-repo querying — --from flag to search issues and knowledge in external repositories
• Section-based editing — --replace-section and --append-to-section for surgical updates
• Bulk import — knowledge import from existing markdown files or design documents
• Auto-injection — Relevant knowledge pages injected into agent context via MCP server
• Conflict resolution — Accept-both merge strategy for concurrent knowledge edits

Configuration Presets

Get the right defaults for your workflow without reading docs.

• Team mode — Strict tracking, required comments, CI verification, enforced commit signing. For shared repos with multiple contributors or agents.
• Solo mode — Relaxed tracking, encouraged comments, local-only verification, signing disabled. For personal projects and solo development.
• Custom — Configure each setting individually via the interactive walkthrough.

# Choose during first-time setup (interactive TUI)
crosslink init

# Or apply a preset directly
crosslink config --preset team
crosslink config --preset solo

# Skip the TUI and use team defaults
crosslink init --defaults

The presets configure tracking strictness, comment discipline, lock stealing policy, kickoff verification level, and signing enforcement. Run crosslink config show to see current settings, or crosslink config --reconfigure to re-run the walkthrough.

Behavioral Hooks & Rules

Your agents follow the rules without being told.

• Issue tracking enforcement — Hooks block code changes without an active crosslink issue
• Comment discipline — Hooks enforce typed comments before commits and issue close
• No-stubs policy — Post-edit hooks detect TODO, FIXME, unimplemented!() stubs
• Drift detection — Adaptive reminders when agent behavior drifts from project norms
• Tracking modes — Strict, normal, and relaxed enforcement (crosslink workflow)
• Language-aware rules — 20+ language-specific rule files auto-injected based on project languages
• House style — crosslink style syncs project conventions from a central git repo
• Local overrides — rules.local/ directory for gitignored per-machine rule customizations

Terminal Dashboard

crosslink tui

Live interactive terminal UI built with ratatui:

• Issues tab — Tree view, detail view, filtering, and sorting
• Agents tab — Active session monitoring with heartbeat status
• Knowledge tab — Page browser with syntax highlighting
• Milestones & Config tabs — Project overview at a glance
• Mouse support, command palette (Ctrl-P), clipboard export, keyboard help (?)

Web Dashboard

crosslink serve

Browser-based dashboard for visual project oversight — charts, drag-and-drop, and real-time agent monitoring.

Maintenance

• crosslink prune — Squash stale hub and knowledge branch history
• crosslink integrity — Data integrity checks (counters, hydration, locks, schema, layout)
• crosslink compact — Manual event compaction

Other

• SSH signing — Agent key generation, per-commit signing, allowed_signers management
• Driver intervention tracking — crosslink intervene logs human corrections for agent improvement
• Typed comments — Comments carry kind (plan, decision, observation, blocker, resolution, result)
• Clock skew detection — Uses git commit timestamps as witness to detect time drift
• Context measurement — crosslink context measures and optimizes context injection overhead
• Lazy auto-hydration — Local database auto-refreshes when the hub branch moves, no manual sync needed
• Config presets — --team and --solo presets for quick setup; layered config with local overrides
• Configurable git remote — Use any remote for hub/knowledge branches, not just origin
• Works everywhere — CLI + VS Code extension + context provider for any AI agent

Installation

Requires Rust 1.87+ (install rustup).

# From crates.io
cargo install crosslink

# From source
git clone https://github.com/forecast-bio/crosslink.git
cd crosslink/crosslink && cargo install --path .

Also available as a VS Code extension.

Documentation

Getting started:

• Your First Agent Session — Get running in under a minute
• Installation — All install methods and requirements

Guides:

• Session Workflow — Persistent memory across conversations
• Multi-Agent Coordination — Distributed locking for parallel AI work
• Design Document Workflow — Interactive, codebase-grounded design authoring
• Kickoff: Autonomous Agents — Launch background agents in worktrees
• Swarm Orchestration — Multi-agent phased builds from design documents
• Container-Based Agents — Run agents in isolated Docker containers
• Knowledge Management — Shared research pages synced via git
• Terminal Dashboard — Interactive TUI for issues and agents
• Web Dashboard — Browser-based project oversight
• Behavioral Hooks — Guardrails for AI coding agents
• Tracking Modes — Strict, normal, and relaxed enforcement
• Maintenance — Pruning, health checks, and database compaction

Reference:

• CLI Reference — Full command reference
• Hook Configuration — Customize enforcement behavior
• Rules Customization — Edit behavioral rules
• Kickoff Report Schema — Agent validation report format

Development

cargo test          # Run tests
cargo clippy        # Lint
cargo fmt           # Format

See also:

• Architecture Overview
• ELI5 Explanation
• API Documentation

License

MIT