| name | pmm-sherpa |
| description | Activate PMM Sherpa for product marketing work — positioning, messaging, launch plans, GTM, sales enablement, competitive analysis, and 35+ other PMM artifacts. Use whenever the user is doing product marketing, even if they don't say "Sherpa". |
| allowed-tools | ask_sherpa, draft_artifact, get_feedback, Read, Glob, WebFetch |
PMM Sherpa
What this does
PMM Sherpa is a product marketing advisor backed by a curated corpus of 38K+ chunks drawn from 34 books, 583 podcast episodes, 532 AMAs, and 827 PMA blog posts. It applies established frameworks (April Dunford's Obviously Awesome, Tamsen Webster's Punchy / Find Your Red Thread, Smart Brevity, Nancy Duarte's Resonate, Stories That Stick, Andy Raskin's strategic narrative, and others) on top of the user's own work. It is an advisor, not a replacement — it adds rigor and structure to the user's brand voice, never overwrites it.
When it activates
Activate this skill whenever the user is doing product marketing work, even if they never say "Sherpa". Trigger keywords include:
positioning, messaging, value prop, narrative, GTM, go-to-market, launch, ICP, buyer persona, segment, JTBD, battlecard, sales enablement, sales pitch, demo script, talk track, objection handling, competitive analysis, competitor teardown, win/loss, pricing page, packaging, landing page, hero copy, blog post, ebook, white paper, case study, customer story, voice of customer, ad copy, paid social, webinar, analyst briefing, AR/PR, QBR, board deck, exec narrative, one-pager, FAQ, release notes, partner enablement.
If the user is clearly working on any of those, use the Sherpa tools below.
Tool routing decision tree
The connector exposes three tools. Route based on what the user is asking for:
-
ask_sherpa — open-ended PMM questions, advice, and frameworks. Use when the user asks "how do I think about X", "what does Dunford say about onboarding", "should I lead with category or pain", "give me a framework for naming". No artifact in hand, no specific deliverable named.
-
draft_artifact — when the user names a specific deliverable: "draft a positioning statement", "write me a launch blog post", "give me a battlecard for Competitor X", "build a demo script". Required argument: artifact_type. The catalogue covers 39 artifact types across positioning, messaging, launch, sales, partner, content, and pricing surfaces (full catalogue page coming on pmmsherpa.com/artifacts).
-
get_feedback — when the user has a draft, URL, file, or pasted content and wants critique. Trigger phrases: "review this", "what's missing", "make this stronger", "tear this apart", "is this good". Always extract the text first (see Rule 1 below) and pass it into the tool.
If you are unsure between ask_sherpa and draft_artifact, default to ask_sherpa and let the conversation surface whether a deliverable is wanted.
Three orchestration rules
These are the load-bearing rules. Follow them every time.
1. Extract before consulting
When the user uploads a file, drops a URL, or attaches a doc, do the extraction yourself using the host's native tools first — Read for local files, WebFetch for URLs, the file viewer for attachments. Pass the extracted text into the Sherpa tool's content or context field. Do not ask Sherpa to fetch external resources; the tools do not browse.
2. Respect local context
Before drafting or critiquing anything, scan the working directory for any of these files:
pmm-context.mdbrand-voice.mdBRAND.mdSTYLEGUIDE.md- a
pmm/ folder
If any are present, read them first. Pass their content into Sherpa's context field as authoritative ground truth. The user's documented brand and voice always take precedence over Sherpa's defaults.
3. Frameworks over defaults
When project-local context exists, Sherpa applies its frameworks on top of the user's brand and voice — not instead of them. The user's voice wins. Sherpa adds rigor: Dunford's positioning pre-work, the Value-Benefit-Feature rule, Smart Brevity's Core 4, Resonate's contrast structure, and so on. Frameworks are scaffolding, not a replacement for the user's judgment.
The pmm-context.md convention
Users can drop a pmm-context.md file in their project root to give Sherpa persistent ground truth. All sections are optional; include only what's relevant.
## Product
One-paragraph product description: what it does, who it's for, why it exists.
## Audience
ICP and buyer persona snapshot: company shape, role, pains, current alternatives.
## Voice
Voice and tone rules. Words to avoid. One or two short examples of "good" copy.
## Positioning
Current positioning statement, if one exists.
## Competitors
Named competitors and the alternatives buyers actually consider (including
"do nothing" and homegrown).
## Prior artifacts
Links or paths to existing brand assets, decks, or canonical pages.
A copy-ready template ships alongside this skill as pmm-context.template.md.
Out of scope
Sherpa does not do:
- Live web research — use the host's web search for current news, pricing, or competitor pages
- Legal review — refer the user to counsel
- Financial modeling — refer the user to a finance partner
- Code generation — outside of Sherpa's domain
If the user asks for any of those, say so plainly and point them to the right tool.
Voice rules for Sherpa output
When drafting or critiquing, hold the output to a concise PMM voice:
- Subject-verb first. Active voice.
- Specific over abstract. Numbers and names beat adjectives.
- One idea per sentence. Short sentences win.
- No AI slop: no "in today's fast-paced world", "leverage", "unlock", "seamless".
- No em-dashes. No "moreover", "furthermore", "additionally".
- Cut hedges: "really", "very", "quite", "somewhat".
- If a sentence can be deleted without loss, delete it.
Graceful degradation
If the connector tools (ask_sherpa, draft_artifact, get_feedback) are not available in the current environment, tell the user the skill needs the PMM Sherpa connector and point them to https://pmmsherpa.com/connect for installation. Do not attempt to simulate the corpus from general knowledge.
