Hiberius

always-on-agent

Use when building a personal AI agent that runs continuously on your own machine - deciding the architecture, waking a laptop that sleeps, choosing a phone channel that will not get a number banned, keeping secrets away from the model, setting approval gates for irreversible actions, or running the read-only first pass that maps projects and credentials before the agent is allowed to touch anything.

Data & Documents Listed

bank-statement-to-table

Use when turning a bank statement into a spreadsheet or a ledger - extracting transactions from a PDF or scan, parsing dates and amounts across European and Anglo conventions, handling multi-line descriptions and page breaks, and above all proving the extraction is correct by checking it against the statement's own running balance before anyone acts on the numbers.

ad-comment-moderation

Use when spam, scam replies, link drops or competitor poaching appear under Facebook or Instagram ads and posts - designing or debugging a comment moderation rule set, deciding what to hide versus keep, finding where comments on a dark post actually live, or building against the Meta Graph API comment endpoints without hiding real customers.

Testing & QA Listed

incrementality-testing

Use when deciding whether an ad, creative, channel or product change actually caused the result rather than took credit for it - designing an A/B or geo test, choosing a sample size, checking a split for sample ratio mismatch, reading a lift that has been watched daily, reducing variance with pre-period data, or converting platform-reported conversions into incremental ones.

Data & Documents Listed

competitor-ad-intelligence

Use when analysing competitors' ads from the Meta Ad Library, Google Ads Transparency Center, TikTok Creative Center or LinkedIn Ads Library - ranking which of their creatives are actually working without spend data, grouping re-uploaded variants so longevity is honest, classifying angles, or finding which angles work for them and are missing from your own account.

DevOps & Infrastructure Listed

lead-delivery-reconciliation

Use when leads are delivered to an affiliate network, lead buyer or CRM and you need to know what was actually accepted and paid rather than what was sent - reconciling a delivery log against a network register, computing acceptance rate and effective payout, normalising or deduplicating leads before delivery, handling rejects and resale, or debugging why a daily cap overshot or a delivered lead never appeared in the buyer's system.

whatsapp-receptionist-builder

Use when building, extending or debugging an AI assistant on WhatsApp that answers customers and books appointments - a receptionist, booking bot or support agent on the WhatsApp Cloud API. Covers webhook signature failures, the 24-hour customer service window, duplicate replies from webhook retries, double-booking under concurrency, voice note handling, human escalation and multi-tenant number routing.

affiliate-tracker-ops

Use when conversions are not attributing, a postback is not firing or is firing twice, a click id stops being forwarded somewhere between the ad and the offer, tracker and platform and network report three different numbers, or a tracking domain, redirect chain or traffic-source macro needs to be set up or debugged. Covers Zeustrack, Voluum, Binom, Keitaro and any tracker built on the same click-id and postback model.

cpa-profit-ops

Use when deciding whether paid traffic actually made money rather than whether it hit a platform metric - reading a CPA or lead-gen report, ranking campaigns, countries, accounts or operators by profit, setting or defending a CPL target, triaging ad accounts that spend without producing, reconciling platform-reported leads against what an affiliate network paid, or working with spend and payouts denominated in different currencies.

Data & Documents Listed

invisible-text-forensics

Use when text might carry hidden Unicode - auditing a document, prompt, CV, contract, dataset or source file for zero-width characters, bidirectional overrides, tag characters, variation-selector payloads or homoglyphs; when a humanizer pass needs to cover more than writing style; when tracing which copy of a document leaked; or when pasted text fails a diff, breaks a regex, or does not match a string that looks identical.