diff --git a/.pi/settings.json b/.pi/settings.json
new file mode 100644
index 0000000..eadb0bc
--- /dev/null
+++ b/.pi/settings.json
@@ -0,0 +1,3 @@
+{
+  "skills": ["/app/skills"]
+}
diff --git a/docker-compose.yml b/docker-compose.yml
index 6341257..f391771 100644
--- a/docker-compose.yml
+++ b/docker-compose.yml
@@ -12,5 +12,7 @@ services:
     volumes:
       # Persistent memory + schedules. Bind-mounted so you can inspect/edit on host.
       - ./memory:/app/memory
-      # Skills are bind-mounted read-only so you can iterate without rebuild.
-      - ./skills:/app/skills:ro
+      # Skills are bind-mounted read-write so the agent can create/update skills.
+      - ./skills:/app/skills
+      # Pi settings + project-level skills discovery
+      - ./.pi:/app/.pi
diff --git a/skills/firecrawl/SKILL.md b/skills/firecrawl/SKILL.md
new file mode 100644
index 0000000..0854497
--- /dev/null
+++ b/skills/firecrawl/SKILL.md
@@ -0,0 +1,302 @@
+---
+name: firecrawl
+description: Web scraping, crawling, searching, and content extraction via Firecrawl API. Use for scraping any URL to clean markdown/HTML/JSON, searching the web with full page content, crawling entire websites, discovering all URLs on a site (map), batch scraping multiple URLs, or AI-powered autonomous data gathering (agent). Requires a Firecrawl API key.
+---
+
+# Firecrawl
+
+Turn websites into LLM-ready data. Scrape, crawl, search, and interact with the web at scale.
+
+- **Base URL:** `https://api.firecrawl.dev/v2`
+- **API Key Format:** `fc-...` (get one at https://firecrawl.dev)
+- **Docs:** https://docs.firecrawl.dev
+
+## Setup
+
+Set your API key as an environment variable before using any endpoint:
+
+```bash
+export FIRECRAWL_API_KEY="fc-YOUR_API_KEY"
+```
+
+If the key is not set, prompt the user to get one at https://firecrawl.dev.
+
+Verify your key works with the helper script:
+
+```bash
+node /app/skills/firecrawl/scripts/firecrawl.js scrape "https://example.com"
+```
+
+## Quick Start — Node.js Helper
+
+A Node.js helper script wraps all endpoints with auto-polling for async jobs:
+
+```bash
+# Scrape a URL to markdown
+node /app/skills/firecrawl/scripts/firecrawl.js scrape "https://example.com"
+
+# Search the web
+node /app/skills/firecrawl/scripts/firecrawl.js search "latest AI news"
+
+# Crawl a website (polls until done)
+node /app/skills/firecrawl/scripts/firecrawl.js crawl "https://docs.example.com"
+
+# Map URLs on a site
+node /app/skills/firecrawl/scripts/firecrawl.js map "https://example.com"
+
+# AI-powered data gathering
+node /app/skills/firecrawl/scripts/firecrawl.js agent "Find the founders of Stripe"
+
+# Batch scrape multiple URLs
+node /app/skills/firecrawl/scripts/firecrawl.js batch "https://a.com" "https://b.com"
+```
+
+Pass options as a JSON string for the second argument:
+
+```bash
+node /app/skills/firecrawl/scripts/firecrawl.js scrape "https://example.com" '{"waitFor":2000}'
+node /app/skills/firecrawl/scripts/firecrawl.js search "AI tools" '{"limit":10}'
+node /app/skills/firecrawl/scripts/firecrawl.js crawl "https://docs.example.com" '{"limit":10}'
+```
+
+## Full API Reference (cURL)
+
+### Scrape — extract content from a single URL
+
+Get LLM-ready markdown, HTML, screenshots, or structured JSON from any page.
+
+```bash
+curl -s -X POST 'https://api.firecrawl.dev/v2/scrape' \
+  -H "Authorization: Bearer $FIRECRAWL_API_KEY" \
+  -H 'Content-Type: application/json' \
+  -d '{
+    "url": "<URL>",
+    "formats": ["markdown"],
+    "onlyMainContent": true,
+    "waitFor": 0
+  }'
+```
+
+**Key parameters:**
+| Param | Type | Description |
+|-------|------|-------------|
+| `url` | string | **Required.** The URL to scrape. |
+| `formats` | array | Output formats: `markdown`, `html`, `rawHtml`, `screenshot`, `screenshot@fullPage`, `json`, `links`, `images`. Use `["markdown"]` for LLM consumption. |
+| `onlyMainContent` | bool | Strip nav, footer, sidebars. Default: `true`. |
+| `waitFor` | int | Time in ms to wait for JS rendering. Use `1000`-`5000` for JS-heavy pages. |
+| `actions` | array | Browser actions before scraping: click, scroll, type, wait. See [Interact](#interact--actions). |
+| `extract` | object | Extract structured data with `{schema: {...}, prompt: "..."}`. |
+| `includeTags` | array | HTML tags to include. |
+| `excludeTags` | array | HTML tags to exclude. |
+| `headers` | object | Custom HTTP headers for the request. |
+
+**Realistic example** — scrape a JS-heavy docs page:
+
+```bash
+curl -s -X POST 'https://api.firecrawl.dev/v2/scrape' \
+  -H "Authorization: Bearer $FIRECRAWL_API_KEY" \
+  -H 'Content-Type: application/json' \
+  -d '{
+    "url": "https://docs.firecrawl.dev/api-reference/v2/scrape",
+    "formats": ["markdown"],
+    "onlyMainContent": true,
+    "waitFor": 2000
+  }' | jq -r '.data.markdown' | head -100
+```
+
+Always use `jq` to extract the relevant fields. The response shape:
+
+```json
+{
+  "success": true,
+  "data": {
+    "markdown": "# Page content...",
+    "html": "...",
+    "metadata": {
+      "title": "Page Title",
+      "sourceURL": "https://...",
+      "statusCode": 200
+    }
+  }
+}
+```
+
+### Search — search the web and get full content from results
+
+```bash
+curl -s -X POST 'https://api.firecrawl.dev/v2/search' \
+  -H "Authorization: Bearer $FIRECRAWL_API_KEY" \
+  -H 'Content-Type: application/json' \
+  -d '{
+    "query": "<search query>",
+    "limit": 5,
+    "scrapeOptions": {
+      "formats": ["markdown"],
+      "onlyMainContent": true
+    }
+  }'
+```
+
+**Key parameters:**
+| Param | Type | Description |
+|-------|------|-------------|
+| `query` | string | **Required.** The search query. |
+| `limit` | int | Max results. Default: 5. |
+| `scrapeOptions` | object | Scrape options applied to each result page (same as scrape params). Omit to get metadata only (no page content). |
+| `sources` | array | Limit search to `["web"]` or `["news"]`. Default: web. |
+
+### Crawl — scrape every page on a website
+
+```bash
+curl -s -X POST 'https://api.firecrawl.dev/v2/crawl' \
+  -H "Authorization: Bearer $FIRECRAWL_API_KEY" \
+  -H 'Content-Type: application/json' \
+  -d '{
+    "url": "<URL>",
+    "limit": 50,
+    "maxDepth": 3,
+    "scrapeOptions": {
+      "formats": ["markdown"],
+      "onlyMainContent": true
+    }
+  }'
+```
+
+Returns a job with an `id`. Poll for status:
+
+```bash
+curl -s -X GET "https://api.firecrawl.dev/v2/crawl/<JOB_ID>" \
+  -H "Authorization: Bearer $FIRECRAWL_API_KEY"
+```
+
+When `status` is `"completed"`, `data` contains the scraped pages array. Crawl jobs are async — poll every 5-10 seconds until done.
+
+**Key parameters:**
+| Param | Type | Description |
+|-------|------|-------------|
+| `url` | string | **Required.** Starting URL. |
+| `limit` | int | Max pages to crawl. Default: 500. |
+| `maxDepth` | int | Max link depth from the starting URL. |
+| `includePaths` | array | Glob patterns to match URLs (e.g., `["/docs/*"]`). |
+| `excludePaths` | array | Glob patterns to exclude URLs. |
+| `scrapeOptions` | object | Scrape options applied to each page. |
+
+### Map — discover all URLs on a site
+
+```bash
+curl -s -X POST 'https://api.firecrawl.dev/v2/map' \
+  -H "Authorization: Bearer $FIRECRAWL_API_KEY" \
+  -H 'Content-Type: application/json' \
+  -d '{"url": "<URL>"}'
+```
+
+**Optional parameters:**
+| Param | Type | Description |
+|-------|------|-------------|
+| `url` | string | **Required.** The URL to map. |
+| `search` | string | Filter links by relevance to a search term. |
+| `ignoreSitemap` | bool | Skip the sitemap and discover links by crawling. |
+| `includeSubdomains` | bool | Include subdomains. |
+
+### Batch Scrape — scrape many URLs at once
+
+```bash
+curl -s -X POST 'https://api.firecrawl.dev/v2/batch/scrape' \
+  -H "Authorization: Bearer $FIRECRAWL_API_KEY" \
+  -H 'Content-Type: application/json' \
+  -d '{
+    "urls": ["<URL1>", "<URL2>", "<URL3>"],
+    "formats": ["markdown"],
+    "onlyMainContent": true
+  }'
+```
+
+Returns a job with an `id`. Poll `GET /v2/batch/scrape/<JOB_ID>` until `status` is `"completed"`.
+
+### Agent — AI-powered autonomous data gathering
+
+Describe what you need — no URLs required. Firecrawl's AI agent searches, navigates, and retrieves data for you.
+
+```bash
+curl -s -X POST 'https://api.firecrawl.dev/v2/agent' \
+  -H "Authorization: Bearer $FIRECRAWL_API_KEY" \
+  -H 'Content-Type: application/json' \
+  -d '{
+    "prompt": "<describe what data you need>",
+    "urls": ["<optional URL to focus on>"],
+    "model": "spark-1-mini",
+    "schema": {}
+  }'
+```
+
+Returns a job with an `id`. Poll `GET /v2/agent/<JOB_ID>` until `status` is `"completed"`.
+
+**Key parameters:**
+| Param | Type | Description |
+|-------|------|-------------|
+| `prompt` | string | **Required.** Natural language description of what data to gather. |
+| `urls` | array | Optional URLs to constrain the search. |
+| `model` | string | `"spark-1-mini"` (default, cheaper) or `"spark-1-pro"` (complex tasks). |
+| `schema` | object | JSON schema for structured output. |
+
+Use `spark-1-pro` when: comparing data across multiple sites, navigating complex/auth'd sites, or when accuracy is critical.
+
+### Interact / Actions — interact with a page before scraping
+
+Use `actions` in a scrape request to click, scroll, type, or wait before extracting content:
+
+```bash
+curl -s -X POST 'https://api.firecrawl.dev/v2/scrape' \
+  -H "Authorization: Bearer $FIRECRAWL_API_KEY" \
+  -H 'Content-Type: application/json' \
+  -d '{
+    "url": "https://example.com",
+    "formats": ["markdown"],
+    "actions": [
+      {"type": "wait", "milliseconds": 2000},
+      {"type": "click", "selector": "button.accept-cookies"},
+      {"type": "write", "selector": "input[name=\"search\"]", "text": "my query"},
+      {"type": "press", "key": "Enter"},
+      {"type": "scroll", "direction": "down", "amount": 500}
+    ]
+  }'
+```
+
+Action types: `wait`, `click`, `write`, `press`, `scroll`, `screenshot`, `executeJavascript`.
+
+## jq Quick Reference
+
+Most responses are JSON. Use `jq` to extract what you need:
+
+```bash
+# Extract markdown from scrape response
+| jq -r '.data.markdown'
+
+# Get search result titles and URLs
+| jq '.data[] | {title: .title, url: .url}'
+
+# Check crawl/agent status
+| jq '{status: .status, completed: .completed, total: .total}'
+
+# Pretty-print anything
+| jq .
+```
+
+## When to Use Each Endpoint
+
+| Task | Endpoint |
+|------|----------|
+| Get content from a known URL | **Scrape** |
+| Find information across the web | **Search** |
+| Scrape an entire website | **Crawl** |
+| List all URLs on a site | **Map** |
+| Scrape a known list of URLs | **Batch Scrape** |
+| "Find me X about Y" (no specific URL) | **Agent** |
+| Page requires login/click/scroll first | **Scrape** with `actions` |
+
+## Cost and Rate Limits
+
+- Each scrape, search result, or crawled page consumes credits based on your plan.
+- Crawl and batch jobs return async — always poll and wait for completion.
+- Check your usage: `GET /v2/team` returns credit and usage info.
+- Always use `onlyMainContent: true` to reduce token waste.
diff --git a/skills/firecrawl/scripts/firecrawl.js b/skills/firecrawl/scripts/firecrawl.js
new file mode 100755
index 0000000..d6ddddb
--- /dev/null
+++ b/skills/firecrawl/scripts/firecrawl.js
@@ -0,0 +1,205 @@
+#!/usr/bin/env node
+/**
+ * Firecrawl CLI helper — wraps all v2 endpoints via Node.js.
+ * Usage: node scripts/firecrawl.js <command> [args...]
+ *
+ * Commands:
+ *   scrape  <url>            Scrape a URL to markdown
+ *   search  <query>          Search the web
+ *   crawl   <url>            Start a crawl, poll until done
+ *   map     <url>            Map all URLs on a site
+ *   agent   <prompt>         AI agent for autonomous data gathering
+ *   batch   <url1> <url2>... Batch scrape multiple URLs
+ */
+
+const API_KEY = process.env.FIRECRAWL_API_KEY;
+const BASE = 'https://api.firecrawl.dev/v2';
+
+if (!API_KEY) {
+  console.error('Error: FIRECRAWL_API_KEY is not set.');
+  console.error('Get one at https://firecrawl.dev and export it:');
+  console.error('  export FIRECRAWL_API_KEY="fc-YOUR_API_KEY"');
+  process.exit(1);
+}
+
+async function request(method, path, body) {
+  const url = BASE + path;
+  const res = await fetch(url, {
+    method,
+    headers: {
+      'Authorization': `Bearer ${API_KEY}`,
+      'Content-Type': 'application/json',
+      'Accept': 'application/json'
+    },
+    body: body ? JSON.stringify(body) : undefined
+  });
+  const text = await res.text();
+  try { return JSON.parse(text); }
+  catch { return text; }
+}
+
+function sleep(ms) { return new Promise(r => setTimeout(r, ms)); }
+
+async function scrape(url, opts = {}) {
+  const body = {
+    url,
+    formats: opts.formats || ['markdown'],
+    onlyMainContent: opts.onlyMainContent !== false,
+    waitFor: opts.waitFor || 0,
+    ...opts
+  };
+  const result = await request('POST', '/scrape', body);
+  if (result.success && result.data?.markdown) {
+    console.log(result.data.markdown);
+  } else {
+    console.log(JSON.stringify(result, null, 2));
+  }
+}
+
+async function search(query, opts = {}) {
+  const body = {
+    query,
+    limit: opts.limit || 5,
+    ...opts
+  };
+  const result = await request('POST', '/search', body);
+  console.log(JSON.stringify(result, null, 2));
+}
+
+async function crawl(url, opts = {}) {
+  const scrapeOpts = { formats: ['markdown'], onlyMainContent: true, ...(opts.scrapeOptions || {}) };
+  const { scrapeOptions, ...restOpts } = opts;
+  const body = { url, limit: 50, scrapeOptions: scrapeOpts, ...restOpts };
+  const job = await request('POST', '/crawl', body);
+  if (!job.id) {
+    console.log(JSON.stringify(job, null, 2));
+    return;
+  }
+  console.error(`Crawl started: ${job.id} — polling...`);
+  for (let i = 0; i < 60; i++) {
+    await sleep(5000);
+    const status = await request('GET', `/crawl/${job.id}`);
+    if (status.status === 'completed') {
+      for (const page of status.data) {
+        console.log(`\n## ${page.metadata?.sourceURL || page.metadata?.url}\n`);
+        console.log(page.markdown || '');
+      }
+      return;
+    }
+    if (status.status === 'failed') {
+      console.error('Crawl failed:', JSON.stringify(status, null, 2));
+      process.exit(1);
+    }
+    console.error(`  ${status.completed || 0}/${status.total || '?'} pages...`);
+  }
+  console.error('Crawl timed out.');
+}
+
+async function map(url, opts = {}) {
+  const result = await request('POST', '/map', { url, ...opts });
+  console.log(JSON.stringify(result, null, 2));
+}
+
+async function agent(prompt, opts = {}) {
+  const body = { prompt, ...opts };
+  const job = await request('POST', '/agent', body);
+  if (!job.id) {
+    console.log(JSON.stringify(job, null, 2));
+    return;
+  }
+  console.error(`Agent started: ${job.id} — polling...`);
+  for (let i = 0; i < 60; i++) {
+    await sleep(5000);
+    const status = await request('GET', `/agent/${job.id}`);
+    if (status.status === 'completed') {
+      console.log(JSON.stringify(status, null, 2));
+      return;
+    }
+    if (status.status === 'failed') {
+      console.error('Agent failed:', JSON.stringify(status, null, 2));
+      process.exit(1);
+    }
+    console.error('  still working...');
+  }
+  console.error('Agent timed out.');
+}
+
+async function batch(urls, opts = {}) {
+  const body = {
+    urls,
+    formats: opts.formats || ['markdown'],
+    onlyMainContent: opts.onlyMainContent !== false,
+    ...opts
+  };
+  const job = await request('POST', '/batch/scrape', body);
+  if (!job.id) {
+    console.log(JSON.stringify(job, null, 2));
+    return;
+  }
+  console.error(`Batch started: ${job.id} — polling...`);
+  for (let i = 0; i < 60; i++) {
+    await sleep(5000);
+    const status = await request('GET', `/batch/scrape/${job.id}`);
+    if (status.status === 'completed') {
+      for (const page of status.data) {
+        console.log(`\n## ${page.metadata?.sourceURL || page.metadata?.url}\n`);
+        console.log(page.markdown || '');
+      }
+      return;
+    }
+    if (status.status === 'failed') {
+      console.error('Batch failed:', JSON.stringify(status, null, 2));
+      process.exit(1);
+    }
+    console.error(`  ${status.completed || 0}/${status.total || '?'} pages...`);
+  }
+  console.error('Batch timed out.');
+}
+
+const [,, cmd, ...args] = process.argv;
+
+async function main() {
+  switch (cmd) {
+    case 'scrape':
+      await scrape(args[0], args[1] ? JSON.parse(args[1]) : {});
+      break;
+    case 'search':
+      await search(args[0], args[1] ? JSON.parse(args[1]) : {});
+      break;
+    case 'crawl':
+      await crawl(args[0], args[1] ? JSON.parse(args[1]) : {});
+      break;
+    case 'map':
+      await map(args[0], args[1] ? JSON.parse(args[1]) : {});
+      break;
+    case 'agent':
+      await agent(args[0], args[1] ? JSON.parse(args[1]) : {});
+      break;
+    case 'batch':
+      await batch(args, args.length > 1 ? JSON.parse(args[args.length - 1]) : {});
+      break;
+    default:
+      console.error(`Usage: firecrawl.js <scrape|search|crawl|map|agent|batch> [...args]
+
+Commands:
+  scrape  <url> [options_json]    Scrape a URL to markdown
+  search  <query> [options_json]  Search the web
+  crawl   <url> [options_json]    Crawl a website
+  map     <url> [options_json]    Map all URLs on a site
+  agent   <prompt> [options_json] AI-powered data gathering
+  batch   <url1> [url2...]        Batch scrape multiple URLs
+
+Examples:
+  node scripts/firecrawl.js scrape "https://example.com" "{\"waitFor\":2000}"
+  node scripts/firecrawl.js search "latest AI news"
+  node scripts/firecrawl.js crawl "https://docs.example.com" "{\"limit\":10}"
+  node scripts/firecrawl.js agent "Find the founders of Stripe"
+  node scripts/firecrawl.js batch "https://a.com" "https://b.com"`);
+      process.exit(1);
+  }
+}
+
+main().catch(e => {
+  console.error('Error:', e.message);
+  process.exit(1);
+});