Put your coding agent on call

Q: Which agents are supported?

Claude Code, Cursor, Windsurf, and anything else that reads Agent Skills. The auto-detection hook is Claude Code–only today; the MCP server and Python SDK cover other setups.

Q: Is it really free?

Yes, 10 root cause analyses a month, no signup. Claim your agent within three days to keep its history. Outgrow it and plans are at steadwing.com/pricing.

Q: Not in a coding agent?

MCP or SDK. Run the MCP server for Claude Desktop and other clients: npx -y @steadwing/mcp-server with a STEADWING_API_KEY from app.steadwing.com. Or pip install steadwing to capture errors from Python services.

Q: Do I have to claim the agent?

No, it expires. Unclaimed agents expire after three days and the skill registers a fresh one next time. Claiming keeps your incident history.

When a production error hits your terminal, Steadwing investigates inside Claude Code, Cursor, or Windsurf, then hands you the root cause in three minutes. No API key to paste. Nothing to deploy.

$ npx skills add steadwing/agent-skills

Free to start: no card, no signup form. Your agent registers itself.

claude — ~/payments-service

$ python3 workers/charge.py --queue prod

Traceback (most recent call last):

File "workers/charge.py", line 45, in process

ConnectionPoolExhausted: pool max=10 reached in 3.2s

▲ steadwing hook production-style error detected

▸ steadwing Run root cause analysis? (y/n) y

✓ Error log + 3 referenced files sent, investigating

▸ claude resuming your task: pagination for /orders…

▸ claude orders endpoint done, tests passing

▸ steadwing Root cause: retry path never releases connections (2m 41s)

→ Full report: app.steadwing.com/incident/7f3a

Sends error data only after you confirm
SOC 2 Type II
Readable, MIT-licensed source

01Install. One command in your terminal, nothing to configure.

02Forget it’s there. In Claude Code, a hook stands watch and flags production-style errors the moment they hit your terminal.

03Next stack trace: confirm, and the root cause lands while your agent keeps coding.

“Couldn’t my agent just debug this itself?”

It can try. And spend your session doing it. Steadwing runs the investigation on its own pipeline while your agent keeps coding.

$ npx skills add steadwing/agent-skills

A pipeline built for incidentsMulti-step analysis with evidence attached, not one long prompt.
A report your team can shareEvery incident becomes a permanent link, not a scroll-back.
Your context window stays yoursAnalysis runs server-side. Your session keeps its memory for the fix.

Nothing leaves your machine until you say yes

The skill is plain markdown. Read it in one sitting. The hook watches locally; sending is always yours to approve. First-run registration sends the agent type, nothing else. Uninstall is one command.

Read the skill before it runs

sent, after you confirm

the error log

up to ten source files the error references

never sent

your repository

.env and credential files

anything you didn’t approve

FAQ

Which agents are supported?

claude code · cursor · windsurf, and anything else that reads Agent Skills. The auto-detection hook is Claude Code–only today; the MCP server and Python SDK cover other setups.

Is it really free?

yes, 10 a month, no signup. Claim your agent within three days to keep its history. Outgrow it and plans are at steadwing.com/pricing.

Not in a coding agent?

mcp or sdk. Run the MCP server for Claude Desktop and other clients: npx -y @steadwing/mcp-server with a STEADWING_API_KEY from app.steadwing.com. Or pip install steadwing to capture errors from Python services.

Do I have to claim the agent?

no, it expires. Unclaimed agents expire after three days and the skill registers a fresh one next time. Claiming keeps your incident history.

Your next error is the demo

$ npx skills add steadwing/agent-skills

Or read it first: docs.steadwing.com/ai-skills