Skills your AI agent can install.

Build procedures from the process that is actually performed, not from an idealized description.

Create a compliant, persuasive proposal without inventing capabilities, credentials, customer results, certifications, pricing, staffing, or legal commitments.

Diagnose why an offer may be misunderstood or unconvincing before rewriting everything.

Transform a source asset into multiple channel-native pieces without changing what the source actually supports.

Create defensible competitor intelligence rather than a table of guesses.

Review an AI-enabled workflow before deployment. Focus on concrete failure paths, permissions, verification, recoverability, and human authorization.

Convert messy meeting notes or transcripts into a reliable operational record.

Determine whether a dataset is fit for the intended decision before analyzing it.

Attack a product requirement before engineering pays for ambiguity.

Use this skill when a user says things like "sales are bad," "we need marketing," "we need more people," "customers keep leaving," "everything takes too long," or proposes a solution before the actual constraint is established.

Variety of useful skills for ChatGPT

Writes Seedance 2.5 / Dreamina prompts: role-mapped references up to 50 materials, 30-second videos built as stages, video editing and forward/backward extension, keyframe sequences, storyboard grids and blockouts, seamless transitions. Includes a mode router and pre-flight checklist.

Writes Seedance 2.0 video prompts: 6-step structure, one-move camera rules, 10 transition types, layered SFX per shot, and a character-reference workflow that holds one face across 15+ shots. Includes a fix table for when a generation misses.

Analyse a subscription pricing model end to end: whether tier boundaries match how value is actually delivered, where upgrades stall, and what a change would do to existing customers.

Review an analytics pipeline for the failures that produce confidently wrong numbers: silent truncation, non-idempotent loads, timezone drift, and joins that quietly multiply rows.

A structured audit of a paid search or social account: wasted spend by device and placement, audience overlap, creative fatigue, and the conversion tracking that decides whether any of it is true.

Run a blameless postmortem that finds the systemic cause instead of the last person to touch it, with a timeline built from evidence and actions that would actually have prevented it.

Diagnose a slow query from its execution plan, propose the smallest effective fix, and predict which rewrites will silently lose their index.

Group keywords by what search results actually show rather than by wording, then design hub-and-spoke pages with an internal link plan and explicit cannibalisation checks.

Write landing copy from evidence you actually have — the product's real behaviour and real customer language — with every claim traceable to something you can point at.

Document an HTTP API from its handlers — real params, real response shapes, real error codes — and flag every place the existing spec has drifted from the code.

Turn a feature or a diff into a test plan organised by risk: what must never break, what is likely to, and which cases are worth automating versus checking once.

Diagnose why an app runs locally but not when deployed: every variable the code reads, cross-checked against each environment, with fail-soft keys called out.

Turn a branch diff into a pull request description a reviewer can act on: what changed, why, what to look at hardest, and what was deliberately left out.