MCPInstall

storytelling-narrative

Structure, refine, or draft narrative content or storytelling for business communication such as slide decks, blog posts, pitches, reports. Use when the user wants help building a narrative arc, find the through-line, sequence beats or refine existing content for flow and impact.

Installs
42
Last 24h
—
Trending rank
—
Security
Not audited

Install

Works with every agent the skills CLI supports.

npx skills add https://github.com/sammcj/agentic-coding --skill storytelling-narrative
  1. 01

    Install

    Run the command in your project. The CLI asks which agents to install for.

  2. 02

    Describe the task

    Ask your agent for something this skill covers. It loads storytelling-narrative automatically when relevant.

  3. 03

    Review the result

    The agent follows the SKILL.md steps, scripts and references bundled with the skill.

SKILL.md

The instructions your agent reads when the skill is triggered.

Storytelling & Narrative

Help the user build, refine, or restructure a narrative for a business artefact: slide deck, blog post, pitch, exec memo, article, or report. The skill is format-agnostic at its core. Format-specific notes live in references/formats.md.

The user is a domain expert. They usually know the material deeply or have rough touchpoints in mind. What they're missing is structure, flow, and a clear through-line. Your job is to help them surface what's already there and shape it, not to invent narrative from nothing or layer marketing language on top.

Operating principles

Lenses, not templates. Storytelling frameworks (Pixar, ABT, Three-Act, StoryBrand, Hero's Journey, Golden Circle, SCQA, Pyramid Principle, Dykes Arc, Duarte Sparkline) are different surface arrangements of the same underlying primitives: status quo, tension, stakes, turn, resolution. Treat named frameworks as lenses to check a story against, not templates to fill in. Pick a lens when it fits the audience-goal-material combination - never force material into a framework that doesn't match.

Narrative comes from the material, not a script. Most business storytelling failures aren't structural - they're the result of not engaging with the material long enough to find what's actually there. Spend real time in Phase 2 (engaging the material) before reaching for any framework.

Anti-slop posture. No marketing adjectives, no hot takes, no cliche verbiage or buzzwords. Concrete nouns over abstract ones. Specific people, specific moments, specific numbers. If a sentence could appear in any company's blog post, rewrite it.

Pause when there's a real fork; proceed when there isn't. Use multi-choice questions (the AskUserQuestion tool) when the user genuinely needs to steer - multiple plausible spines, multiple framework lenses, multiple orderings, output-mode choice. Don't pause on small calls or things you can infer with confidence. The user has explicitly said: increase autonomy when you have understanding, ask when you don't.

Narrative primitives

Every narrative - regardless of framework - is built from these. Find them in the material before structuring anything.

  • Status quo / "what is" - the world the audience already lives in. Familiar enough that they nod.
  • Tension - what's wrong, missing, or about to break. The reason the story exists.
  • Stakes - what happens if nothing changes. Without stakes, the audience has no reason to care.
  • Turn - the moment, decision, evidence, or insight that changes things. The most often-missing primitive in weak business narratives.
  • Resolution - the new state, the action, the call. What the audience now does, believes, or decides.
  • Hook - the entry point that opens the loop. Must land within the first 30-60 seconds (or first paragraph). Six common types in references/hooks-and-moments.md.
  • Through-line - a single sentence the rest of the piece serves: "from X to Y because of Z". If you can't write the through-line, the narrative isn't ready.

The shape is always: old understanding → tension → new understanding → new action. Different frameworks just arrange this differently.

Workflow

This is a phased flow, not a rigid sequence. Skip phases the user has already done; double back when new information surfaces. The pause points below are where multi-choice user input genuinely steers the work.

Phase 1: Get oriented

Don't start structuring until you understand audience, goal, material state, and format. This isn't a "quick diagnostic" - these answers shape everything downstream. Ask substantively, but with focus.

Things you need to know:

  • Audience. Who is this for? What do they already know? What do they care about? What position or fatigue are they bringing in?
  • Goal. What should the audience think, feel, or do after consuming this? A specific decision, a belief shift, an action, an emotional response?
  • Material state. Raw knowledge in their head? Rough touchpoints sketched out? An existing draft to refine? An existing deck/data to reshape?
  • Format and constraints. Slide deck, blog post, pitch, memo, article, report? Length? Time-to-deliver? Live presentation or read-alone?
  • Tone. Visionary / pragmatic / friendly / urgent / credible / sceptical / confident.

Use AskUserQuestion for things with discrete plausible answers (tone, audience type, format, material state). Use open prose questions when you need substance (audience-specific concerns, goal nuance).

If the user volunteered most of this in their initial prompt, don't re-ask - confirm what you understood and ask only what's missing.

Phase 2: Engage with the material

This is the most undervalued phase. Skip it and you'll produce a structure that doesn't fit what the user actually has to say.

Branch by material state:

Raw knowledge / rough touchpoints - ask the user to talk through what they have. Don't summarise back yet - keep extracting. Probe for narrative substrate:

  • "What changed? What made you start paying attention to this?"
  • "What was the moment you realised X?"
  • "Who's affected? What do they actually do differently?"
  • "What's the part most people get wrong?"
  • "What number or specific example surprised you?"

You're looking for moments, not abstractions. Concrete things - a meeting, a number, a customer, a failure, a near-miss - are the raw material narratives are built from.

Existing draft - read carefully. Find the implicit arc (or note its absence). Mark:

  • Where the hook is (or where it should be)
  • Whether there's a turn - and where
  • What's the spine the rest serves
  • What's repeated, what's buried, what's missing

Read references/critique.md for the diagnostic checklist.

Existing data / deck - look for the transformation already present. The story is usually there waiting to be uncovered: a before-after, a problem-solution, a counterintuitive finding, a shift in trend. Resist the urge to invent a story; surface the one that's there.

Output of this phase: a list of raw narrative material - concrete moments, facts, characters, tensions, changes - not yet ordered.

Phase 3: Find the spine

Distill the raw material to:

  • A single-sentence through-line: "from X to Y because of Z"
  • A candidate hook
  • The turn - what shifts
  • The stakes - why it matters if nothing changes
  • The resolution - new state, new action

If you can find more than one plausible spine, that's a fork - present 2-3 options to the user via AskUserQuestion and let them pick. Different spines lead to genuinely different narratives; this is one of the highest-impact decision points.

If there's only one spine that fits the material, propose it directly with rationale. Don't manufacture options for the sake of asking.

Phase 4: Choose a structure (lens)

Match a framework lens to the audience-goal-material combination. See references/narrative-frameworks.md for the full catalogue and selection guidance. Quick orientation:

  • ABT (And/But/Therefore) - short pieces: exec updates, elevator pitches, paragraph openers, slide subtitles.
  • Three-Act - general default for blog posts, talks, longer writing.
  • Pixar (Once upon a time...) - change stories, adoption stories, "how X became Y".
  • Golden Circle (Why/How/What) - vision, purpose, mission, manifesto.
  • StoryBrand SB7 - customer-facing marketing where the customer is the hero.
  • Hero's Journey (condensed) - founder origin, brand history, personal narrative.
  • Duarte Sparkline (what is / what could be) - persuasive keynotes, change comms.
  • Dykes Data Storytelling Arc - data-driven presentations and insight reports.
  • SCQA (Situation/Complication/Question/Answer) - exec briefings, consulting memos.
  • Pyramid Principle (Minto) - top-down written reports for executives who scan.

Lenses are not exclusive - a piece can use SCQA at the document level and ABT at the paragraph level. Present 2-3 lens options when more than one would work; explain the trade-offs briefly; let the user pick.

Phase 5: Order the beats

Arrange the material into the chosen structure. Three things to attend to:

  • Sequence for emotional and logical flow. The audience should always know why they're reading the next paragraph. Each beat earns the next.
  • Cut ruthlessly. Most business narratives carry 30-50% material that doesn't serve the through-line. If a beat doesn't move the audience from where they were to where they need to be, it goes. Mark cuts as "park for later" rather than deleting outright - the user may reuse them elsewhere.
  • Place the turn. The turn is usually mid-to-late, not upfront. Resist "executive summary" instinct that puts the conclusion first - that kills the narrative tension. Exception: read-alone formats for time-poor execs (memos, reports) often do need bottom-line-up-front; in that case, give a one-sentence headline then tell the story.

Show the user the proposed ordering. Pause via AskUserQuestion only if there are multiple genuinely plausible orderings - otherwise propose and let them react.

Phase 6: Decide output mode

Before drafting anything, ask the user what level of help they want. Use AskUserQuestion with options like:

  • Structure only - they take it from here and write the prose themselves.
  • Hook + key moments - draft the opener and the most load-bearing beats; they fill the rest.
  • Full draft - draft prose for every beat in their voice.
  • Refine existing draft to new structure - rework what they already wrote against the new spine.

This is non-negotiable as a pause point. Don't assume the level of involvement the user wants.

Phase 7: Draft (if requested)

When drafting, hold to:

  • Concrete over abstract. Specific nouns, specific numbers, specific people. "The auth team rewrote the session middleware in two weeks" beats "The team transformed our authentication".
  • Verbs do the work. Strong verbs over verb+adjective combos.
  • Vary sentence length. Mix short with long. Don't write five sentences of the same length and structure in a row.
  • The user's voice, not yours. Match the tone they signalled in Phase 1 and the rhythm of any existing material they shared.
  • Anti-slop check. No marketing adjectives, no AI tells, no "Additionally" / "Furthermore" / "It's worth noting" sentence openers, no closing summaries that say "In conclusion, this approach...".
  • Show stakes and consequences. Don't gloss over the tension - that's where the narrative energy lives.

For format-specific drafting concerns (slide one-idea-per-slide rule, blog opener strength, pitch tightness), read references/formats.md.

Phase 8: Self-review

Before returning the final output, run a critique pass against references/critique.md. Common failure patterns:

  • Hook is generic or buried
  • Turn is missing or under-emphasised
  • Stakes implied but not stated
  • Through-line drifts mid-piece
  • Slop language crept in during drafting

Fix these before delivering. If you find structural issues, surface them rather than silently rewriting - the user may have specific intent behind a choice that looks weak.

When to use which reference

  • references/narrative-frameworks.md - when picking a lens, or when the user wants to commit to a specific framework. Selection guidance, per-framework failure modes, and detail on the lesser-known lenses.
  • references/hooks-and-moments.md - when crafting a hook or sharpening a turn. Six hook types, the turn test, placement principles.
  • references/formats.md - when format-specific concerns matter (slide deck rules, blog opener, pitch arc length, memo bottom-line-up-front).
  • references/critique.md - when reviewing existing content, or running self-review on drafted output.

Multi-choice question patterns

The user has explicitly asked you to use AskUserQuestion where it earns its place. Heuristics for when to pause:

Pause forDon't pause for
Tone (visionary / pragmatic / urgent / credible) when not signalledWord-level choices
Audience type when ambiguousTiny edits
Multiple plausible spines (genuine fork)Obvious cuts
Framework lens with 2-3 viable fitsSingle-fit lens
Ordering when multiple orderings are validMechanical sequencing
Hook variant when 2-3 are equally plausibleWhen one hook clearly fits

Always pause for output mode (structure-only / hook + key moments / full draft / refine existing draft) before drafting. Don't infer what level of help the user wants.

Ask 1-3 questions at a time, not 5+. Let the user steer; don't quiz them.

Gotchas

  • Don't fall back to the same handful of frameworks every time. Pixar, Three-Act, Golden Circle, StoryBrand, Hero's Journey, ABT are widely known, but they're a subset. SCQA, Pyramid Principle, Dykes Data Storytelling Arc, and Duarte's Sparkline are often the better fit for exec, consulting, data-driven, or report-style work. Pick by audience-goal-material fit, not by familiarity.
  • This skill is narrative-first and format-agnostic. Charts, dashboards, and visual design are not in scope. If the piece is primarily data-driven visuals, hand off to the storytelling-with-data skill; otherwise stay on the narrative.
  • Don't skip Phase 2. Engaging with the material is what makes the rest work. Frameworks applied to thin material produce slop.
  • Don't manufacture options to ask the user. If there's one obvious answer, propose it. Multi-choice questions are for genuine forks, not for performance of consultation.
  • Don't lecture the user about narrative theory. They want help with their content, not a lesson. Apply the lens silently; surface only what they need to make decisions.
  • Don't add closing "In summary..." paragraphs to drafted content. That's a slop tell. The narrative should land naturally on its resolution.
  • The turn is the most under-built primitive in business writing. Most weak drafts have a status quo, stakes, and a vague resolution but no clear shift. Watch for this when reviewing existing content.

FAQ

What does the storytelling-narrative skill do?

Structure, refine, or draft narrative content or storytelling for business communication such as slide decks, blog posts, pitches, reports. Use when the user wants help building a narrative arc, find the through-line, sequence beats or refine existing content for flow and impact.

How do I install storytelling-narrative?

Run npx skills add https://github.com/sammcj/agentic-coding --skill storytelling-narrative in your project. The skills CLI detects the agents you use and copies the skill into the right folder. For Claude Code you can also place the skill folder at ~/.claude/skills/storytelling-narrative/ (personal) or .claude/skills/storytelling-narrative/ (project).

Which AI agents can use storytelling-narrative?

Any agent that supports the open Agent Skills format (a folder with a SKILL.md file), including Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI, Windsurf, Cline, OpenCode and more.

Is storytelling-narrative safe to install?

No partner security audit is available yet. Skills can run commands and scripts, so read the SKILL.md and any bundled files before installing.

Who maintains storytelling-narrative?

It is published by sammcj in sammcj/agentic-coding. It has 42 recorded installs.

More from sammcj/agentic-coding

View all →

Critically analyse the reasoning in written content (articles, blogs, transcripts, reports), not code. Use only when explicitly asked to critically analyse or critique an argument.

AI & Agents2.4K installs

Applies the Diataxis framework to create or improve technical documentation. Use when being asked to write high quality tutorials, how-to guides, reference docs, or explanations, when reviewing documentation quality, or when deciding what type of documentation to create. Helps identify documentation types using the action/cognition and acquisition/application dimensions.

Docs & Writing615 installs

creating-development-plans

sammcj/agentic-coding

Creates structured development plans with phased task breakdowns, requirements, and QA checklists. Use when the user explicitly asks to create a dev plan, development plan, or document development requirements.

ghostty-config

sammcj/agentic-coding

Open the skill to read its SKILL.md instructions.

AI & Agents221 installs

code-simplification

sammcj/agentic-coding

Open the skill to read its SKILL.md instructions.

AI & Agents203 installs

prompt-enhancer

sammcj/agentic-coding

Open the skill to read its SKILL.md instructions.

AI & Agents184 installs

Popular ai & agents skills

View category →

google-agents-cli-publish

google/agents-cli

This skill should be used when the user wants to "publish an agent", "publish my ADK agent", "register an agent with Gemini", "publish to Gemini" (formerly Gemini Enterprise), or needs guidance on the agents-cli publish gemini command. Also use when the user wants to "manage agents in Agent Registry", "list/update/delete registered agents", or "register an MCP server". Covers ADK vs A2A registration modes, programmatic and interactive usage, flag reference, auto-detection from deployment metadata, Agent Registry fleet management, and troubleshooting. Part of the agents-cli skills suite. Do NOT use for deployment (use google-agents-cli-deploy).

AI & Agents513.9K installs

google-agents-cli-eval

google/agents-cli

This skill should be used when the user wants to "run an evaluation", "evaluate my agent", "evaluate my ADK agent", "write an eval dataset", "analyze eval failures", "compare eval results", "optimize agent", or needs guidance on the Gemini platform eval methodology and the Quality Flywheel. Covers eval metrics, dataset schema, LLM-as-judge scoring, and common failure causes. Applies to any agents-cli project, whatever framework the agent is written in. Do NOT use for agent API code patterns (ADK: use google-agents-cli-adk-code), deployment (use google-agents-cli-deploy), or project scaffolding (use google-agents-cli-scaffold).

AI & Agents513.5K installs

skill-creator

anthropics/skills

Create new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy.

OfficialAI & Agents416.4K installs

codex-pet

prime-skills/runcomfy-agent-skills

Codex Pet generator on RunComfy. Build a Codex-compatible Codex Pet spritesheet.webp + pet.json from a single reference image, drop it into ${CODEXHOME:-$HOME/.codex}/pets/ / and Codex picks it up as a custom Codex Pet next to the 8 built-ins. This skill produces the exact Codex Pet atlas Codex expects (1536x1872 PNG/WebP, 8 cols x 9 rows, 192x208 cells, 9 animation states — idle, running-right, running-left, waving, jumping, failed, waiting, running, review). Calls OpenAI GPT Image 2 edit ONCE via the local RunComfy CLI as runcomfy run openai/gpt-image-2/edit to produce a canonical Codex Pet pose, then assembles all 9 animation rows programmatically with ImageMagick micro-transforms — no Codex Pro, no $imagegen, no OPENAIAPIKEY required, only RUNCOMFYTOKEN. Triggers on "codex pet", "create codex pet", "make codex pet", "hatch codex pet", "/hatch image", "desktop pet codex", "codex pets", "spritesheet.webp", or any explicit ask to build a custom pet for OpenAI Codex.

AI & Agents402.1K installs

full-output-enforcement

leonxlnx/taste-skill

Overrides default LLM truncation behavior. Enforces complete code generation, bans placeholder patterns, and handles token-limit splits cleanly. Apply to any task requiring exhaustive, unabridged output.

AI & Agents401K installs

executing-plans

obra/superpowers

Use when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available

AI & Agents245.7K installs