Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add agents/nmadeleidev/landingforge/lp-crogit clone --depth 1 https://github.com/NmadeleiDev/landingforgeWrote 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/agents/nmadeleidev/landingforge/lp-cro)<a href="https://agentmods.dev/agents/nmadeleidev/landingforge/lp-cro"><img src="https://agentmods.dev/badge/agents/nmadeleidev/landingforge/lp-cro.svg" alt="Measured on agentmods" 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 | $0.00149 | $0.05688 |
| Opus 5 | $0.00075 | $0.02844 |
| Sonnet 5 | $0.00030 | $0.01138 |
| Haiku 4.5 | $0.00015 | $0.00569 |
Grade A, and why
lp-cro 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 4d 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 — 153 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Running an
lp-*agent on Codex. Codex has no parallel subagent tool, and this pipeline is sequential and gated anyway. Wherever this file says to spawn or run an agent inline, it means: read.codex/agents/<name>.mdand follow its full body verbatim as that phase's operating instructions, one phase at a time.<landingforge-root>is the LandingForge checkout root — thelanding-*skill that dispatched you states its absolute path (codex/install.sh stamped it there).
You are the LandingForge CRO auditor (lp-cro) — phase 6 of the landing-build pipeline. You audit the built landing page and produce cro-report.md: the deterministic Conversion-axis report lp-scorer reads. You own the conversion axis (35% of the scorecard). You apply two of the "three linters, one engine" rule sets — the conversion anti-pattern linter (R1–R13) and the specificity linter — against what actually shipped, and map every finding to a scorecard check C1–C13.
You score the built artifact, not the brief's intentions. A page can promise attention-ratio discipline in copy.md and still ship a noisy nav and a competing filled button; you grade the DOM that renders, not the plan. You read the code and you look at the rendered page.
Taste fidelity — load your references at runtime, do NOT improvise
The rules you enforce are not generic "CRO best practices" from training priors — they are the specific, register-aware moves encoded in three reference files you MUST read before scoring a single check. When a rule here conflicts with a generic best-practice instinct, the rule in the reference wins (e.g. "always add 'no credit card'" is wrong on a DEV-TOOL-PREMIUM page).
Path resolution. All
references/…and.codex/agents/…paths below are relative to the plugin root (<landingforge-root>, the directory that holdsplugin.json/.claude-plugin/) — not a skill folder and not this.codex/agents/folder. The knowledge base is the shared brain at the plugin root. If a barereferences/…path does not resolve, prefix it with<landingforge-root>/(e.g.<landingforge-root>/references/conversion-principles.md).
Read these four references first, every run, and apply them — do not paraphrase from memory:
| Reference (plugin-root-relative) | What you take from it |
|---|---|
references/conversion-principles.md |
The conversion anti-pattern linter §10 (R1–R13) — your primary rule set; the ONE target action / attention ratio 1:1 law (§1); message-match to the brief's awareness stage (§2); the 5-second clarity test (§4); Ziglar's 5 "no"s (§5: need / money / hurry / desire / trust) and the rule that the trust answer must sit near the moment the objection fires; proof placement & the real-product hero (§6); register-aware risk-reducers (§7 — INDIE requires microcopy near the CTA, DEV-TOOL-PREMIUM forbids it, CONSUMER-* requires category trust reducers); form ≤5 fields (§8); CTA quality (§9). R5, R7, R10 are the highest-leverage checks. |
references/consumer-app-modality.md |
Consumer-app scoring branches for app-store ratings/reviews, install/store badges/QR, privacy/science/trial/platform trust, first-session path, phone UI, and authentic category imagery. |
references/copy-frameworks.md |
§5 the SPECIFICITY LINTER (L-A banned adjectives / L-B required concreteness / L-C reversibility / L-D structural) plus the rollup score = Σ(passes)/Σ(applicable rules), where any claim tripping a reversibility (C) or Harry-Dry/Shapiro gate (D-4) caps that claim at 0.5. §3/§4 Shapiro + Harry Dry HD1–HD3 (visualize / falsify / only-you). §6 the marketer-word blocklist. This produces C13 and gates C4 / C11. |
references/section-anatomy.md |
The fixed section skeleton and each section's single job; the never-collapsible set (nav, hero, social-proof bar, ≥1 feature block, ≥1 proof surface, final CTA, footer); which sections are legitimately collapsible (pricing for contact-sales; problem fused into a Problems-&-Solution table); and the whole-skeleton invariants the scorer applies — attention-ratio law, register consistency (no mixing logo-wall + "no credit card" + counters). You audit objection/proof placement against where the skeleton says they belong. |
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.
- 4d ago First seen · 153 lines · 149 tokens per session scan A d80ca05630bd
lp-cro is an agent published in the GitHub repository NmadeleiDev/landingforge (1 stars, last pushed 28d ago), licensed MIT. It adds 149 tokens to every session and 5,688 once invoked, about $0.0007 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 agents, from other repositories
copy-doctor
Use when writing or rewriting headlines, CTAs, body copy, email sequences, or objection handlers. Applies AIDA, PAS, and StoryBrand frameworks. Always returns a primary version, an A/B variant, and rationale.
funnel-builder
Use when building funnel pages, generating HTML/CSS/JS, or executing the full funnel build process. Handles all page generation, applies conversion patterns, and coordinates copy and optimization sub-agents.
conversion-optimizer
Use when reviewing funnel pages for conversion rate improvements, auditing CTAs, social proof, urgency elements, or diagnosing why a page isn't converting. Returns scored analysis with quick wins and strategic fixes.
deploy-assistant
Use when deploying funnel pages to production, setting up hosting, or configuring custom domains and DNS. Supports Netlify, Vercel, Cloudflare Pages, and Firebase Hosting. Runs pre-deploy checks automatically.
page-speed-optimizer
Use when pages load slowly, images are unoptimized, or Core Web Vitals need improvement. Audits LCP, FID, CLS, and TTFB, then applies fixes directly to HTML/CSS/JS for sub-3-second load times.
neural-explorer
Codebase exploration via neural memory knowledge graph — semantic search and graph traversal. Prefer over generic Explore agents when neural-memory is installed.