feat: chrome extension + AI content capture pipeline with question bank

.kiro/skills/interview-extractor.md

@@ -0,0 +1,241 @@
+---
+name: interview-extractor
+description: >
+  Extracts and reformats a frontend interview experience from a URL or raw
+  pasted text into a clean, structured, third-person markdown case study.
+  Works universally — no external APIs required. Just give it a link or text.
+triggers:
+  - 'extract interview'
+  - 'format this interview'
+  - 'rewrite this experience'
+  - 'process this article'
+  - 'structure this interview'
+  - 'convert to interview format'
+  - 'parse interview experience'
+---
+
+# Interview Experience Extractor Skill
+
+## Purpose
+
+Transform raw interview experience content (from a URL or pasted text) into a
+clean, structured, third-person case study that is genuinely useful for
+candidates preparing for frontend engineering interviews.
+
+---
+
+## Step-by-Step Instructions
+
+When this skill is triggered, follow these steps in order:
+
+### Step 1 — Acquire the content
+
+- **If a URL is given**: Fetch the full text content of the page. Strip HTML
+  tags, navigation, footers, ads, and any non-article boilerplate. Keep only
+  the article body.
+- **If raw text is given**: Use it directly. Strip any Medium/DEV/Hashnode
+  header noise (author name, read-time badge, clap counts, "Follow" buttons).
+- If neither is provided, ask: "Please share either a URL or paste the raw
+  article text."
+
+### Step 2 — Assess quality
+
+Score the content 1–10 for being a **genuine frontend/web interview experience
+narrative** using this rubric:
+
+| Score | Meaning                                                                                            |
+| ----- | -------------------------------------------------------------------------------------------------- |
+| 9–10  | Detailed narrative: specific rounds, verbatim/paraphrased questions, company named, outcome stated |
+| 7–8   | Good narrative but missing some details (e.g. outcome unknown, fewer questions)                    |
+| 5–6   | Mostly tips/guide format — not a personal experience                                               |
+| 1–4   | Not a frontend interview experience at all                                                         |
+
+- If score < 7, respond: "This doesn't appear to be a detailed frontend
+  interview experience (score: X/10). Here's why: [brief reason]. I can still
+  attempt a basic summary if you'd like."
+- If score ≥ 7, proceed to Step 3.
+
+### Step 3 — Extract metadata
+
+Identify from the content:
+
+- `company_name` — the company being interviewed at
+- `role` — job title (e.g. "Senior Software Engineer — Frontend")
+- `level` — one of: intern / junior / mid / senior / staff
+- `outcome` — one of: selected / rejected / pending / unknown
+- `location` — city, remote, or hybrid if mentioned
+- `author` — name of the person who wrote the experience (if available)
+- `source_platform` — where it was originally published (Medium, DEV,
+  Hashnode, LinkedIn, etc.) derived from the URL or article metadata
+- `ctc` — compensation details if mentioned, else omit
+- `rounds_count` — total number of interview rounds
+
+### Step 4 — Rewrite in strict third person
+
+**CRITICAL RULE**: The entire output MUST be in **strict third person**.
+
+| ❌ WRONG (first person)       | ✅ CORRECT (third person)                           |
+| ----------------------------- | --------------------------------------------------- |
+| "I was asked about closures"  | "The candidate was asked about JavaScript closures" |
+| "My first round was DSA"      | "The first round focused on DSA"                    |
+| "I solved it using a hashmap" | "The candidate approached it using a hashmap"       |
+| "I got the offer"             | "The candidate received an offer"                   |
+
+**NDA / No questions disclosed**: If the article explicitly states an NDA and
+gives no questions, acknowledge this in the round breakdown and focus on
+round structure, skills tested, and difficulty instead of specific questions.
+
+### Step 5 — Output the structured markdown
+
+Produce the following sections in this exact order. Do not skip any section.
+If data is unavailable for a section, write "Not mentioned" rather than
+omitting the section.
+
+---
+
+## Output Format
+
+```markdown
+## Overview
+
+[2–3 sentences: company, role, total rounds, outcome. Third person only.]
+
+## Role & Compensation Details
+
+- **Position**: [role title]
+- **Company**: [company name]
+- **Location**: [city / remote / hybrid]
+- **Level**: [intern/junior/mid/senior/staff]
+- **Experience Required**: [years if mentioned, else "Not mentioned"]
+- **CTC / Stipend**: [if mentioned, else omit this line]
+- **Outcome**: [Selected ✅ / Rejected ❌ / Pending ⏳ / Unknown]
+
+## Interview Process Summary
+
+[Bullet list of all rounds in order — name, type, duration if available]
+
+- Round 1: [Name] — [Type] — [~duration if known]
+- Round 2: ...
+
+## Round-by-Round Breakdown
+
+### Round N — [Round Name]
+
+**Type**: [coding / machine-coding / system-design / conceptual / behavioral / hr]
+**Difficulty**: [Easy / Medium / Hard]
+
+**Questions asked:**
+
+1. [Question verbatim or paraphrased — mark "(paraphrased)" if not exact]
+2. ...
+
+**What the interviewer focused on:**
+[What skills/depth were being evaluated]
+
+**Tips for this round:**
+[1–2 specific, actionable tips derived from this experience]
+
+---
+
+[Repeat for each round]
+
+## Key Technical Topics Covered
+
+- **JavaScript / TypeScript**: [specific topics: closures, promises, generics…]
+- **React / Framework**: [hooks, rendering, state management…]
+- **CSS / Layout**: [flexbox, grid, animations…]
+- **System Design**: [component design, API design, caching…]
+- **DSA**: [data structures and algorithms covered]
+- **Behavioral**: [themes: conflict, leadership, growth…]
+
+## Preparation Tips
+
+1. [Concrete, actionable tip specific to this company/role]
+2. [Another tip]
+3. [Another tip]
+4. [Another tip — optional]
+5. [Another tip — optional]
+
+## Verdict
+
+[2–3 sentences: outcome, overall difficulty rating (1–5 stars), whether the
+candidate recommends the process/company, and any standout observation about
+the hiring culture.]
+```
+
+---
+
+## Metadata Block
+
+After the markdown, output this JSON block (fenced) for programmatic use:
+
+```json
+{
+  "company_name": "string",
+  "role": "string",
+  "level": "intern|junior|mid|senior|staff",
+  "outcome": "selected|rejected|pending|unknown",
+  "difficulty": 1-5,
+  "quality_score": 1-10,
+  "rounds": [
+    {
+      "name": "string",
+      "type": "coding|machine-coding|system-design|conceptual|behavioral|hr",
+      "difficulty": "easy|medium|hard"
+    }
+  ],
+  "topics": ["react", "javascript", "..."],
+  "questions": [
+    {
+      "question": "string",
+      "type": "machine-coding|dsa|system-design|conceptual|behavioral",
+      "difficulty": "easy|medium|hard",
+      "topics": ["string"]
+    }
+  ],
+  "source_platform": "Medium|DEV|Hashnode|LinkedIn|other",
+  "suggested_slug": "kebab-case-seo-slug-max-8-words"
+}
+```
+
+---
+
+## Edge Case Handling
+
+| Situation                       | How to handle                                                                                                    |
+| ------------------------------- | ---------------------------------------------------------------------------------------------------------------- |
+| NDA — no questions disclosed    | Focus on round structure, skills tested, difficulty. Note "Questions not disclosed (NDA)" in each round section. |
+| First person throughout         | Rewrite entirely in third person. Do not leave any "I/my/me/we" references.                                      |
+| No clear outcome                | Set outcome to "unknown". Do not guess.                                                                          |
+| Multiple companies mentioned    | Extract the PRIMARY company being interviewed at.                                                                |
+| Tips/guide format (score 5–6)   | Offer a partial extraction with a note that it's a guide, not a narrative.                                       |
+| Very short content (<500 words) | Note that detail is limited and some sections may be incomplete.                                                 |
+| Paywalled or inaccessible URL   | Ask the user to paste the text directly.                                                                         |
+
+---
+
+## Example Trigger
+
+User says:
+
+> "Extract interview: https://medium.com/@user/google-frontend-interview-2024"
+
+or:
+
+> "Format this interview experience: [pastes raw text]"
+
+The skill then follows Steps 1–5 and outputs the full structured markdown +
+metadata JSON.
+
+---
+
+## Quality Checklist (self-verify before outputting)
+
+Before producing the final output, verify:
+
+- [ ] No first-person pronouns anywhere in the markdown
+- [ ] Every round has a difficulty tag
+- [ ] Preparation Tips are specific to this experience, not generic
+- [ ] Metadata JSON is valid and all required fields are present
+- [ ] Suggested slug is ≤ 8 words, kebab-case, SEO-friendly
+- [ ] Source platform is derived from URL, not from the `source` field in DB