Pmm Context

PMM Context

What this skill does

This is the foundation file every other PMM skill reads before doing any work. Without it, you'll get generic output. With it, every output is grounded in the user's actual product, market, ICP, and voice.

When to invoke this skill

• The user wants the agent to retain structured knowledge about their company or product.
• A sibling skill (positioning, messaging, personas, customer-research, launch) is about to run and no context file exists yet.
• The user asks to update, refresh, or audit their context.

Where the context file lives

Check, in order:

1. pmm-context.md in the current working directory or attached files
2. .agents/pmm-context.md
3. .claude/pmm-context.md
4. The user's project or workspace knowledge (search for "pmm-context")

If none exist, offer to create one together.

Workflow

When creating a new context file

1. Tell the user this will take about 10 minutes and produces a file every other skill will reference. Ask if they have an existing one-pager, deck, or website to start from — if so, ingest it first and ask only the gaps.

1. Walk through the sections below conversationally, one section at a time. Don't dump the whole questionnaire at once. After each section, summarize what you heard back to the user before moving on.

1. When all sections are filled, output the complete markdown file and ask the user to save it as pmm-context.md (or in the agent's project knowledge store).

When updating an existing context file

Read the existing file in full. Then ask: "What's changed since this was written?" Update only the affected sections. Show a diff-style summary of what changed.

Required sections

The output file must follow this exact structure so other skills can rely on the headings:

# PMM Context: [Company Name]

_Last updated: YYYY-MM-DD_

## Product
- **What it is:** [one sentence]
- **Category:** [e.g., "AEO/GEO infrastructure", "API observability"]
- **Stage:** [pre-seed | seed | Series A | Series B+ | public]
- **Pricing model:** [subscription | usage-based | per-seat | freemium | enterprise]
- **Core capabilities:** [3-7 bullets]

## ICP (Ideal Customer Profile)
- **Company type:** [e.g., "B2B SaaS, 50-500 employees, mid-market"]
- **Buyer titles:** [primary and secondary]
- **Geographies:** [primary markets]
- **Trigger events:** [what makes them start looking — e.g., "new CMO joins", "AI strategy mandate"]

## Personas
[For each persona, a brief profile. Full personas live in their own files generated by the `personas` skill.]
- **[Persona name]:** [1-line description, primary pain, primary gain]

## Positioning
- **Market category:** [the box you're putting yourself in]
- **Primary competitive alternatives:** [what customers do today — not just direct competitors]
- **Unique attributes:** [2-4 things only you have]
- **Value (so what?):** [the benefit those attributes enable]
- **Best-fit customers:** [the segment for whom value is strongest]

## Competitors
| Name | Type | Their positioning | Where we beat them | Where they beat us |
|---|---|---|---|---|

## Voice & tone
- **Voice attributes:** [3-5 adjectives, e.g., "sharp, plainspoken, technical-but-not-jargony"]
- **Words/phrases we use:** [house style — e.g., "PMM not 'product marketer'"]
- **Words/phrases we avoid:** [e.g., "synergy", "best-in-class", "leverage as a verb"]
- **Audience reading level:** [e.g., "informed practitioner, not novice"]

## Proof points
- **Customer logos we can name:** [list]
- **Case studies / quotes:** [link or paste 1-3 representative ones]
- **Metrics we cite:** [e.g., "40% reduction in review cycles"]

## What we don't do (and won't claim)
[Important guardrails — what the product isn't, who isn't a fit. Helps the model refuse to overclaim.]

How other skills use this file

Other skills in this repo (positioning, adaptive-messaging, personas, customer-research, launch, aeo, outreach, and sales-enablement) look for pmm-context.md first, then layer their work on top. If it is missing, they can invoke this skill or collect a smaller task-specific brief and continue.

Quality bar

• If a section is unknown, mark it _TBD_ rather than guessing. Other skills will surface those gaps.
• Keep sections short and specific. The file should be 1-3 pages, not 20. Long context dilutes attention.
• Use the user's actual language for product names and category, not the model's paraphrase.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: dirknicol
• Source: dirknicol/pmm-skills
• License: MIT

Install and usage instructions live in the source repository linked above.