#58
Fix CSV export on large files
Feedback hub bug tied to the export reliability release.
AI agents
Give your coding agent the whole picture.
Your agent starts every session knowing nothing about your plan. Connect Frostbyte and it reads the task, the release it belongs to, and the feedback behind it, then writes progress back while you build.
Selected task
1.5.0
Target: Jan 15, 2026
Export reliability and feedback hub fixes for the next public release.
8
Tasks
3
2
Open
1
In Progress
4
5
Done
Agent handoff
Claude Code
v2.1.216
Found FB-58. It affects large exports, came from Feedback Hub, and belongs to v1.5.0.
Moving FB-58 to In Progress
Updating api-csv.ts to stream rows
Edited 2 files+57-3
Ran the CSV export test path and verified the large-file case.
PR ready. FB-58 marked done
Works with Claude Code, Codex, Cursor, and any other MCP client.
Used by developers at
Both sides of the loop
One agent in your terminal, one in the app.
Two agents, two jobs. Your coding agent works from the plan; the built-in Frostbyte Agent keeps the workspace itself in order.
01
Your coding agent
Claude Code, Codex, or Cursor connects over MCP. It reads tasks, releases, and signals, moves work as it builds, and captures what users reported along the way.
02
The Frostbyte Agent
A chat panel inside every project. Ask what needs attention, have it tidy the board or draft a release, and approve anything destructive before it happens.
What you get
A tracker your agent can read and write.
No pasting task descriptions into prompts, no updating the board afterwards. The workspace is the context.
01
Hosted MCP server
Point any MCP client at getfrostbyte.dev/mcp with an API token. Nothing to install, no local process to keep running.
02
One-command plugin
One command in Claude Code or Codex wires up MCP, skills, and slash commands for the full workflow.
03
17 tools
List, create, and update projects, tasks, areas, releases, and signals without leaving the session.
04
Progress written back
The agent moves tasks and updates release progress as it works. Open boards update in real time.
05
Signals from the terminal
Mention that a user hit a bug and the agent captures it as a signal, waiting in your review queue.
06
Approval gates
The in-app agent asks before anything destructive. Deletes and declines need an explicit yes from you.
How it works
Connect once, then just work.
01
Connect your agent.
Install the Frostbyte plugin in Claude Code or Codex, or add the hosted MCP server to any client with an API token from your settings.
02
Ask what's next.
The agent reads the active release and the tasks in it, along with the feedback each one came from. No pasted context, no stale Markdown plan.
03
Build.
It moves the task to In Progress, does the work in your repo, and records a summary of what it did on the task.
04
Everything else follows.
The task closes, release progress ticks up, and anything users reported during the session is sitting in Signals for your next review.
FAQ
Common questions about agents.
Which agents work with Frostbyte?
The plugin supports Claude Code and Codex. The hosted MCP server works with Cursor and any other MCP client, so you're not locked to one vendor.
Which plan do I need?
The MCP server and plugin are Indie and above. The in-app Frostbyte Agent and other AI features work on every plan through your monthly credit pool.
Does connecting my agent use AI credits?
No. Your coding agent runs on its own model and Frostbyte just serves it context over MCP. Credits are only used by AI features inside the app, like the Frostbyte Agent and release notes drafts.
Can an agent delete my data?
The in-app agent has approval gates: destructive actions like deleting tasks need an explicit yes, and every write is membership-checked. Your coding agent goes through its own permission prompts before calling any tool.
Close the loop
Put your agent in the loop.
No credit card. Free stays free.
Connect Claude Code or Codex with one command.