Getting it into your agent
It runs from inside its repository, so the clone comes first — what it calls does not travel with the file alone.
git clone --depth 1 https://github.com/prashishh/seo-geo-report-enginenpx agentmods add skills/prashishh/seo-geo-report-engine/web-researchWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/prashishh/seo-geo-report-engine/web-research)<a href="https://agentmods.dev/skills/prashishh/seo-geo-report-engine/web-research"><img src="https://agentmods.dev/badge/skills/prashishh/seo-geo-report-engine/web-research/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/prashishh/seo-geo-report-engine/web-research"><img src="https://agentmods.dev/badge/skills/prashishh/seo-geo-report-engine/web-research.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00187 | $0.01350 |
| Opus 5 | $0.00093 | $0.00675 |
| Sonnet 5 | $0.00037 | $0.00270 |
| Haiku 4.5 | $0.00019 | $0.00135 |
Grade A, and why
web-research scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 10d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 89 lines — stays where its author put it; the contents beside it link to each section on GitHub.
web-research
Live, cited web research via OpenRouter (Perplexity's Sonar model family), for the moments
static WebFetch/WebSearch aren't enough: multi-hop synthesis across many sources, or
literally asking an AI engine what it currently says about a brand or topic.
Why this exists
WebFetch reads one page; WebSearch lists results. Neither synthesizes across sources with
citations, and neither answers "what does an AI answer engine say about us right now" — which
is itself a live GEO signal, complementary to Ahrefs Brand Radar (and useful when Brand Radar or
the Ahrefs MCP is unavailable, as happens with limited API budgets).
Tool
Stdlib, no install. tools/connectors/openrouter.py:
python3 tools/connectors/openrouter.py probe "What platform do hotels use for cashless tipping?"
python3 tools/connectors/openrouter.py deep "voice of customer for hotel digital tipping products"
Or from Python: from connectors.openrouter import deep_research, probe. Both return
{"answer", "sources": [{"url","title","date"}], "citations": [url, ...], "model", "usage"} —
the connector normalizes OpenRouter's actual response shape (citations arrive as
message.annotations[].url_citation, not the top-level citations field Perplexity's own docs
describe) so callers don't need to know which shape came back. Needs OPENROUTER_API_KEY in
config/secrets.env (get one at openrouter.ai/keys; pay-as-you-go, no separate Perplexity account).
Two modes — pick deliberately, cost differs ~50-100x
- Deep research (
deep_research(),perplexity/sonar-deep-research, ~$0.04-0.05/call). An autonomous multi-step search-and-reason agent. Use for:- Voice-of-customer mining (
customer-research) when Reddit/forums block direct crawling — ask it to synthesize what people say, with sources, instead of fetching threads one by one. - Competitor fact-gathering (
competitor-analysis,comparison-pages) — verified, sourced facts about a rival instead of manually fetching each vendor page. - Market sizing statistics (
market-opportunity) that need a cited, current figure.
- Voice-of-customer mining (
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 10d ago First seen · 89 lines · 187 tokens per session scan A 8bf284772d70
web-research is a skill published in the GitHub repository prashishh/seo-geo-report-engine (5 stars, last pushed 2mo ago), licensed MIT. It adds 187 tokens to every session and 1,350 once invoked, about $0.0009 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
content-amplifier
Use when the user asks to "amplify influencer content with paid media", "set up whitelisting or Spark Ads", "decide which posts to boost", "repurpose influencer content", "turn one video into multiple ads", or "build a UGC asset library"; produces (paid mode) a content-selection scorecard, a paid amplification…
campaign-planner
Use when the user asks to "plan an influencer campaign", "build a campaign blueprint", "track or close a creator campaign", or "record a late campaign correction"; produces the plan and, when requested, a non-canonical evidence tracker with scoped identity, publication, reconciliation, close, and reopen receipts. Not…
cold-outbound-sequencer
Use when the user asks to "build a B2B cold-outbound sequence", "design reply-triage branching", "plan a domain warmup / sending throttle", or "make my outbound CAN-SPAM / opt-in compliant"; produces a multi-step outbound sequence with reply-triage branches (positive / objection / referral / not-now / opt-out), a…
inbox-placement-monitor
Use when the user asks to "track where my emails are actually landing after I send", "read my seed-list inbox vs spam vs promotions results", "trend my Gmail Postmaster / Microsoft SNDS reputation", or "did placement drop after my last send"; produces a per-provider inbox/spam/promotions placement read, a domain/IP…
subject-line-lab
Use when the user asks to "generate subject line variants", "pre-score my subject lines", or "will this subject get truncated / trigger spam filters"; produces a labeled subject + preheader variant set and a per-variant heuristic pre-score card — spam-trigger flags, length/truncation across desktop + mobile…
email-sequence-designer
Use when the user asks to "design a welcome flow", "set up an abandoned-cart sequence", "build a light re-engagement branch inside a lifecycle flow", or "plan a cold-outbound sequence"; produces general lifecycle automation flows (welcome, cart, browse-abandon, post-purchase, in-flow re-engagement, B2B cold outbound)…