aakashg/pm-claude-skills

# Idea Validator

Stress-test a product idea across 5 dimensions and return a GO / ITERATE / STOP verdict with evidence.

## Step 0 — Read first

| Source | Path | What to extract |
|--------|------|-----------------|
| Project context | `CLAUDE.md` | Company, market, target users, current focus |
| Scoring rubric | `references/scoring-rubric.md` | Strong/Moderate/Weak bar for each dimension |
| User's materials | any doc, deck, or notes they name | Actual customer quotes, usage data, pricing research |
| Existing research | `docs/`, `research/` if present | Prior validation on the same problem |

If the user names no materials, say so explicitly in the verdict. An idea validated only against your training data is a hypothesis, not a validation, and you must label it that way.

# LinkedIn Post Writer

Turn a rough idea into one publishable LinkedIn post: one hook, one format, one CTA.

## Step 0 — Read first

Read these before writing a single line. Do not write from memory of "how LinkedIn posts go."

| Source | Path | What to extract |
|--------|------|-----------------|
| Project context | `CLAUDE.md` | Who the user is, their product, banned words, audience |
| Post patterns | `references/post-patterns.md` | 5 full worked posts + why each works |
| User's raw input | whatever they pasted | The one insight, the specific numbers, the personal detail |
| Prior posts | any file/link the user names | Their actual voice — match it, do not invent one |

# Product Design Reviewer

Review a design across 6 dimensions and return prioritized, actionable feedback.

## Step 0 — Read first

| Source | Path | What to extract |
|--------|------|-----------------|
| The design | screenshot, image, or Figma link the user shares | The actual screen — never review from a text description alone |
| Project context | `CLAUDE.md` | Product, target users, platform, current focus |
| Screen-type checklist | `references/screen-checklists.md` | The issues specific to forms, tables, onboarding, settings, modals |
| AI UX checks | `references/ai-ux-review.md` | Extra checks when the design includes AI features |
| Prior feedback | any doc the user names | What was already flagged — do not re-flag resolved issues |

# Prompt Engineer

Diagnose a prompt, rewrite it, and show exactly what changed and why.

## Step 0 — Read first

| Source | Path | What to extract |
|--------|------|-----------------|
| The prompt | whatever the user pasted | Actual wording — never paraphrase before diagnosing |
| Failing output | the output they got, if provided | The failure mode; this determines the fix |
| Project context | `CLAUDE.md` | Audience, product, banned words, output preferences |
| Technique reference | `references/techniques.md` | Full before/after examples for each technique |

If the user pasted a prompt but no failing output, ask for one example of what it produced. Diagnosing from the prompt alone guesses at the failure mode.

# Status Update Writer

Turn messy notes into a precise stakeholder update that stands alone without a follow-up meeting.

## Step 0 — Read first

| Source | Path | What to extract |
|--------|------|-----------------|
| User's raw notes | whatever they pasted | What actually shipped, real numbers, real dates, real names |
| Project context | `CLAUDE.md` | North star metrics, OKRs, company terminology, audience norms |
| Cadence formats | `references/cadences.md` | Daily / monthly / QBR variants of the base format |
| Last update | any prior update they name | What was promised last week — the delta is the story |
| Metrics source | dashboard link or data they provide | Current values only. Never estimate a metric. |