Browser Research Lite

API-key-free online research via the built-in browser tool. Use this skill when web_search fails due to missing keys, and you still need web evidence.

Security Scan

The skill's code, runtime instructions, and requested access largely match its stated purpose (checking/using a local browser tool for web retrieval); no signs of hidden network exfiltration or unrelated credential access were found, though there are a few operational notes to review.

Purpose & Capability

Name/description (API-key-free browser research) align with the included code: index.js invokes a local Python guard and returns guidance for manual/agent-driven browser use. The files focus on detecting a locally-available OpenClaw browser node and do not request unrelated cloud credentials or services.

Instruction Scope

SKILL.md directs running a browser guard script and using the built-in browser tool; the guard script (scripts/browser_guard.py) reads recent OpenClaw session log files and invokes the local 'openclaw browser status --json' command to determine availability. Reading session logs is within the stated purpose (to infer browser availability), but those logs may contain prior user queries or state — the user should be aware the script parses local session files. Also SKILL.md includes a path 'skills/browser-research-lite/scripts/browser_guard.py' that does not match the repository layout (script actually at scripts/browser_guard.py), an operational inconsistency to fix.

Install Mechanism

No install specification and no third-party network downloads; the skill is instruction/code-only and runs local scripts. That is low-risk from an installer perspective.

Credentials

The skill declares no required env vars or credentials, and indeed does not ask for API keys. However, the guard script accesses files under ~/.openclaw/agents/main/sessions (user home) and executes the local 'openclaw' CLI. That is proportionate for detecting a local browser node, but it does give the skill read access to session logs which may contain sensitive conversational context.

Persistence & Privilege

The skill does not request always:true, does not modify other skill configs, and contains no install-time persistence. It only runs local checks and returns status; normal autonomous invocation is permitted by default but not flagged here.

Assessment

This skill appears to do what it says: check whether a local OpenClaw browser node is available and then guide browsing-based retrieval. Before installing, note two things: (1) the guard script reads recent session files under ~/.openclaw/agents/main/sessions and runs the local 'openclaw' CLI — those session logs can contain prior conversation context, so only install if you trust the skill/source and are comfortable with local log reads; (2) SKILL.md references a script path that doesn't match the repo layout (an operational bug you may need to fix). If you want extra assurance, inspect scripts/browser_guard.py yourself and verify it only prints JSON locally (it does) and does not send data externally. If you don't want local session logs to be read, do not install or modify the script to restrict file access.

Goal: perform reliable web retrieval without Brave/Tavily/Serper API keys.

When To Use

• web_search reports key/config errors (for example: missing_brave_api_key).
• You need factual lookup, definitions, theorem statements, or reference pages.
• Benchmark tasks need external evidence but remote search APIs are unavailable.

Core Policy

• Use local computation tools first for computable questions.
• If online retrieval is required, use the built-in browser tool directly.
• Prefer trusted sources (official docs, textbooks, university pages, Wikipedia as secondary).
• Cross-check at least 2 sources for non-trivial claims.

Browser Workflow

1. Run browser availability guard first:

python3 skills/browser-research-lite/scripts/browser_guard.py

2. If browser_available=false, stop browser retries and switch to local tools.
   • If browser_running=false or browser_cdp_ready=false, attach browser manually:
     • open any page in Chrome with OpenClaw extension installed,
     • click the OpenClaw extension icon once to attach current tab,
     • rerun browser_guard.py.
3. If browser_available=true, open a search page with concise query terms.
4. Scan top results and open 2-3 high-quality sources.
5. Extract only the minimal facts needed to answer the question.
6. If pages are noisy, refine query with exact keywords and retry once.
7. Produce final answer with concise rationale; avoid copying long passages.

Fallbacks

• If browsing is blocked/captcha-heavy, switch to alternative domains and shorter queries.
• If browser node is unavailable, avoid repeated browser calls in the same question.
• If browser remains unavailable, switch to skills/web-fetch-research-lite/SKILL.md and run URL-first retrieval via web_fetch.
• If no reliable source is found quickly, return best-effort answer and mark uncertainty.

Solidify Note

When this skill improves benchmark progress, record:

• trigger signal(s),
• source quality and retrieval steps,
• measurable effect on benchmark score/accuracy.