Skip to main content
skill CX Codex

steel-reliability — Claude Code skill: install and use

This skill helps investigate unreliable Steel browser automation involving CAPTCHAs, blocks, proxies, profiles, and retries. Use it when login or session behavior breaks down.

Install into Claude Code · Skills CLI (Vercel) All methods and platforms ↓
npx skills add steel-dev/skills --skill steel-reliability -a claude-code
Open on GitHub

What steel-reliability does

steel-reliability gives an agent a fact-led way to investigate Steel browser workflows that stop behaving consistently. It covers CAPTCHA loops, 403 and human-check pages, proxy failures, profile or identity issues, lost login state, suspicious request pacing, redirect loops, and retry behavior.

The workflow starts with steel doctor --preflight, then checks for a failed session ID, screenshots, logs, and error text. It preserves evidence before configuration changes, checks profiles, pacing, concurrency, proxy or CAPTCHA access, and reruns the smallest failing step. Supporting material covers the mitigation ladder, feature gates, proxies, CAPTCHAs, credentials, identity, pacing, and failure signals.

On Hobby plans, Steel-managed useProxy and CAPTCHA solving should not be enabled silently; feature access comes first. When a failed session has no evidence yet, use steel-session-debugging before this skill.

What it helps with

  • trace a CAPTCHA loop or 403 response
  • inspect proxy and profile failures
  • preserve evidence from a broken session
  • adjust pacing, retries, or concurrency
  • check access to proxy and CAPTCHA features

Example prompts for the agent

  • Investigate this Steel session failure with a CAPTCHA loop and propose the least invasive next step.
  • Run steel doctor --preflight, then explain why login state and the profile are not persisting.
  • Review these redirect and retry failures, preserving logs and screenshots before changing configuration.

How to install steel-reliability

1Skills CLI (Vercel)

npx skills add steel-dev/skills --skill steel-reliability -a claude-code

2Manually: put the files in place

~/.claude/skills/steel-reliability/SKILL.md

Copy the whole skill folder; for one project use .claude/skills/steel-reliability/

Step by step, with commands for each platform: How to install skills in Claude Code: five ways

Questions about steel-reliability

How do I install steel-reliability in Claude Code?

Run npx skills add steel-dev/skills --skill steel-reliability -a claude-code, then use ~/.claude/skills/steel-reliability/SKILL.md.

What does steel-reliability do?

It guides diagnosis and mitigation for unreliable Steel automation involving blocks, CAPTCHAs, proxies, profiles, login state, and retries. The workflow favors the smallest justified change and verifies it against the failing step.

Can Hobby plans use Steel-managed proxies or CAPTCHA solving?

Do not silently turn on Steel-managed useProxy or CAPTCHA solving. Check feature access first; otherwise use defaults, BYOP when a proxy is supplied, profiles and pacing, or ask the user to upgrade.

Source text

SKILL.md · MIT
---
name: steel-reliability
description: Use this skill when a Steel automation appears unreliable because of bot detection, CAPTCHA loops, 403/access-denied or verify-you-are-human pages, proxy failures, profile or identity problems, credential or login persistence issues, suspicious pacing, redirect loops, or retry strategy, or when the user asks about stealth, proxies, CAPTCHA solving, profiles, credentials, or reliable logins. Do not use as a first step for unknown failed sessions; collect evidence with steel-session-debugging first when needed.
license: MIT
compatibility: claude-code,codex,cursor,opencode,pi
metadata:
  owner: steel
  category: reliability
  stage: beta
---

# Steel Reliability

## Skill Boundary

Use this skill to plan and verify legitimate reliability improvements for Steel browser workflows.

Do not promise guaranteed bypass. Do not provide advice for evading access controls or violating site terms. Start with the least invasive change that fits the evidence.

If the user has a failed session but no evidence yet, load `steel-session-debugging` first.

## Setup Check

1. Confirm `steel doctor --preflight` passes.
2. Confirm whether the user has a failed session ID, screenshot, logs, or error text.
3. Check feature access before enabling paid CAPTCHA or Steel-managed proxy features.
4. Preserve evidence before changing configuration.

## Mitigation Ladder

1. Determine whether the failure is normal auth/app behavior or likely anti-bot/reliability.
2. Preserve evidence: status codes, redirects, challenge text, screenshots, console errors, failed requests.
3. Use profiles when login state, cookies, or reputation matter.
4. Add realistic pacing and lower concurrency.
5. Add Steel-managed proxies only when feature access supports them, or BYOP when the user supplies a proxy.
6. Add CAPTCHA solving when feature access supports it and the target flow allows it.
7. Tune stealth config only with clear justification.
8. Verify by rerunning the smallest failing step.

## Feature Check

```bash
curl -sSfL "https://api.steel.dev/v1/details" \
  -H "steel-api-key: $STEEL_API_KEY" | jq -r '.plan'
```

On Hobby plans, do not silently enable Steel-managed `useProxy` or CAPTCHA solving. Use defaults, BYOP when supplied, profiles, pacing, or ask the user to upgrade.

## References

- `references/mitigation-ladder.md`: staged diagnostic and mitigation workflow.
- `references/stealth-and-antibot.md`: challenge signals and acceptable framing.
- `references/feature-access.md`: plan checks and feature gates.
- `references/proxies.md`: Steel-managed proxies, geotargeting, and BYOP.
- `references/captchas.md`: detection, status, manual solve, and failure handling.
- `references/profiles-and-identity.md`: persistent identity and session context.
- `references/credentials.md`: secure login injection patterns.
- `references/pacing-and-behavior.md`: rate, wait, retry, and concurrency guidance.
- `references/failure-signals.md`: what evidence suggests each reliability class.
- `references/session-debugging-handoff.md`: when to collect more evidence first.

© steel-dev · MIT · steel-dev/skills

Similar skills

All skills for Browser and scraping →

From the same set2

Project sponsors

Companies that keep this analytics open
Gold sponsor

Impressions, clicks and code copies — open statistics. We keep the numbers public for the same reason NeBlask itself is public. Become a sponsor