Web Research

Web Research

• Default: Moth search discovery; Moth browser fetch + verify.
• Discovery: never trust search snippets alone; always consider them as unverified leads.
• Fetch: fetch page content with moth browser, consider it the verification source.
• Social: for Meta social domains and Reddit, do not fetch page text or HTML; screenshots only.
• Media: for YouTube, PDF text, X, and podcasts, use the dedicated Moth commands as-is; transcribe YouTube videos and podcasts when useful.

moth tools for Readiness

• Doctor: run moth tools doctor --pretty before workflows that need external tools.
• Install: when doctor reports a missing trusted tool that blocks the task, install it; ask approval for barely adopted tools.

moth search for Web Search

• Scope: discovery only.
• Mode: prefer moth search web "..." --pretty.
• Images: use moth search images "..." --pretty when image discovery helps.
• Videos: use moth search videos "..." --pretty when video discovery helps.
• Output: use --output when artifacts help.
• Limits: use --count, --max-results, --offset, --lang, --country, and --safe when useful.

moth browser for Web Fetch

• Fetch: full pages with moth browser.
• Rate: Moth serializes browser work by registrable domain, but does not enforce 1 req/sec or hourly quotas; self-limit to 1 req/sec/domain and 30–100 req/hour/domain.
• Backoff: Moth API clients retry 429 with Retry-After/exponential backoff; treat 403 as stop/manual backoff, not automatic retry.
• Commands: prefer moth browser open|ax-tree|metadata|screenshot|pdf|wait|click|input|download.
• Shell: always quote vars.
• Validate: URLs, selectors, output paths.
• Artifacts: for moth browser screenshot|pdf, always write to a mktemp-generated path under the OS temp directory; never rely on default filenames; do not write capture files to cwd or repo paths unless I ask.
• State: default to local project browser state only when needed; otherwise avoid persistent sessions.
• Sessions: never mix authenticated sessions with untrusted sites.
• Commit: never commit browser session state or browser capture artifacts.
• Close: always close the browser when you're done with a prompt.

Social Domains

• Scope: Meta social media domains and Reddit.
• Domains: facebook.com, instagram.com, threads.net, reddit.com, old.reddit.com, redd.it.
• Rule: no text fetch, no HTML fetch, no metadata extraction for content verification.
• Allowed: moth browser screenshot only, with temp output paths.

Dedicated Moth Sources

• YouTube: use moth youtube search|metadata|subtitles|audio as-is; transcribe with moth youtube audio, then moth transcribe.
• PDF: use moth pdf2txt as-is.
• X: use moth x search|post|user|user-lookup as-is.
• Podcast: use moth podcast search|episodes|audio as-is; transcribe with moth podcast audio, then moth transcribe.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: alnah
• Source: alnah/agent
• License: Apache-2.0
• Homepage: https://alnah.me

Install and usage instructions live in the source repository linked above.