Tracer: Solving Context Loss in AI Coding Agents

October 15, 2025

If you've worked with AI coding agents like Claude, Cursor, or Copilot, you know the drill: the agent forgets what it was working on between sessions. You end up repeating context, manually tracking tasks in scattered markdown files, and hoping the AI doesn't lose the thread.

I built Tracer to solve this.

The Problem: Context is Fragile

AI agents are incredible at writing code, but they have a flaw: they don't remember. Each new session starts fresh. Your carefully structured TODO list in README.md? The agent might see it, might not, or might misinterpret it.

Markdown TODOs break down fast:

  • No structure. Is "Fix login" blocked by database work? Who knows.
  • No persistence. The agent can't reliably update its own task list.
  • No discovery. How do you find what's actually ready to work on?
  • No history. When did this task change? Why was it closed?

    • The Solution: Structured, Dependency-Aware Tracking

    Tracer is a CLI-first issue tracker built for AI agents. Humans can use it too, but the design choices are all about making agents effective across sessions.

    1. Dependencies Matter

    Not all tasks are equal. Some block others. Tracer tracks this explicitly:

    # Create tasks
tracer create "Design database schema" -t task
tracer create "Build login API" -t task

# Link them: API depends on schema
tracer dep add task-2 task-1 --type blocks

# See what's ready to work on RIGHT NOW
tracer ready
# → Shows task-1 (schema design)
# → Hides task-2 (blocked by task-1)

    2. AI-Native Design

    Every command has a --json flag. Every change is logged. Every operation is fast (around 5ms). The storage format is JSONL, which is both human-readable and machine-parseable.

    # Agent can parse this output programmatically
tracer list --status open --json

    3. Git is the Database

    Tracer stores everything in .trace/issues.jsonl. One line per issue.

    What this gives you:

  • Clean git diffs. You can see exactly what changed.
  • No external database or server needed.
  • Works offline.
  • Syncs across machines via git.
  • Full version history for free.

    • Clone the repo, you get the entire issue database. Push changes, everyone stays in sync.

    Why This Matters

    AI agents need tools built for how they actually work - not human tools retrofitted for machines. That's what Tracer is. Structured over free-form. Programmatic over GUI. Distributed over centralized. Fast over feature-rich.

    It's not trying to be Jira or Linear. It's built to make AI agents useful for complex, multi-session work.

    Try It

    cargo install --git https://github.com/Abil-Shrestha/tracer
tracer init
tracer learn  # Interactive tutorial
    Tracer Demo

    It's open source, MIT licensed, and built with Rust. If you're building with AI agents, try it out. If something's broken, tell me what.

    ⭐ Star on GitHub • Report Issues

    Back to home