x-keyword-comment

# X — Keyword Comment

> keyword + reply intent → search X tweets → read tweet content → generate contextual replies → post to reply area

## Language

All process output to user (progress updates, process notifications) follows the user's language.

## Objective

Search X by keyword, read each tweet's content, generate contextual replies based on a configured brand persona, and post them — all within a browser-act session.

## Prerequisites

- `config/keyword-comment-config.json` has been filled in with actual product, persona, and tone values (all `YOUR_*` placeholders replaced before first run)

## Session Rule

`{SESSION}` is a temporary, per-run session name used in all `browser-act --session {SESSION}` commands below. It is generated at execution start (e.g., `xkc-{timestamp}`) and not persisted across runs.

## Pre-execution Checks

### 1. Tool Readiness

If browser-act has been confirmed available in the current conversation → skip this step.

Invoke `browser-act` via Skill tool to load usage. If installation or configuration issues arise, follow its guidance to resolve then retry.

### 2. Load Config

```bash
python -c "
import json, pathlib, sys
for base in ['.claude/skills/x-keyword-comment', 'output/x-keyword-comment']:
    cfg = pathlib.Path(base) / 'config/keyword-comment-config.json'
    if cfg.exists():
        print(json.dumps(json.loads(cfg.read_text(encoding='utf-8')), ensure_ascii=False, indent=2))
        sys.exit(0)
print('ERROR: config/keyword-comment-config.json not found', file=sys.stderr)
sys.exit(1)
"
```

Hold `product.*`, `persona.*`, `tone.*` fields in working memory for reply composition.

### 3. Browser Selection

List available browsers:

```bash
browser-act browser list
```

- If browsers exist → present the list to the user and let them choose which browser to use for this X session.
- If no browsers exist → guide the user to create one (e.g., `browser-act browser create --type stealth --headed`), then repeat the list step.

Once the user selects a browser, record its ID as `{BROWSER_ID}` for this run.

### 4. Open Session

Generate a unique session name (e.g., `xkc-{timestamp}`) as `{SESSION}`. Open the browser:

```bash
browser-act --session {SESSION} browser open {BROWSER_ID} https://x.com/ --headed
```

If the browser is already open with an active session, list sessions and reuse:

```bash
browser-act session list
```

Pick the session associated with `{BROWSER_ID}` and assign its name to `{SESSION}`.

### 5. Login Verification

If X login status has been confirmed in the current conversation → skip this step.

Otherwise: `browser-act --session {SESSION} get markdown` and check:
- Sidebar bottom shows `@username`, top navigation shows Home / Explore → logged in, continue
- Page shows a "Sign in" button with no logout entry → not logged in; inform the user that login is required and assist the login flow

User refuses or cannot log in → terminate execution.

## Capability Components

> This Skill's operational boundary = what the user can manually do in their browser. It only reads data already displayed to the logged-in user, never bypassing authentication or access controls. JS code is encapsulated in Python files under `scripts/`, invoked via `browser-act --session {SESSION} eval "$(python scripts/xxx.py {params})"`. `$(...)` is bash syntax; use the bash tool for execution.

Below are all atomic capabilities discovered and verified during the exploration phase, listed by command template with parameters. Simply invoke them as needed — no need to read `scripts/*.py` source code or re-verify. Only inspect scripts when execution fails for troubleshooting. Combine freely as needed during execution.

### AI Workflow: Pre-reply Warmup

Warm up the account before posting replies to simulate organic browsing behavior.

**Skip condition**: warmup already performed today and less than 4 hours ago, or user says "fast mode".

**Step 1 — Check notifications and messages (2–3 min)**

```bash
browser-act --session {SESSION} navigate "https://x.com/notifications"
browser-act --session {SESSION} wait stable
browser-act --session {SESSION} get markdown
sleep $((RANDOM % 31 + 60))   # 60–90 s
browser-act --session {SESSION} navigate "https://x.com/messages"
browser-act --session {SESSION} wait stable
browser-act --session {SESSION} get markdown
sleep $((RANDOM % 31 + 30))   # 30–60 s
```

**Step 2 — Browse feed and like (3–5 min)**

```bash
browser-act --session {SESSION} navigate "https://x.com/home"
browser-act --session {SESSION} wait stable
browser-act --session {SESSION} get markdown
```

Randomly pick 3–5 tweets from the feed. For each:

```bash
browser-act --session {SESSION} navigate "{tweet URL}"
browser-act --session {SESSION} wait stable
sleep $((RANDOM % 26 + 15))   # 15–40 s
# If content is relevant → like it:
browser-act --session {SESSION} state
browser-act --session {SESSION} click {Heart index}   # element with aria-label containing "Like"
browser-act --session {SESSION} wait stable
sleep $((RANDOM % 8 + 8))     # 8–15 s
browser-act --session {SESSION} navigate "https://x.com/home"
sleep $((RANDOM % 16 + 10))   # 10–25 s
```

Target: like 1–3 tweets; daily cap **20–30 likes** (avoid fast bulk likes that trigger rate limits).

**Step 3 — Keyword search browsing (2–3 min)**

```bash
browser-act --session {SESSION} navigate "https://x.com/search?q={KEYWORD_ENCODED}&f=live"
browser-act --session {SESSION} wait stable
browser-act --session {SESSION} get markdown
```

Open 2–3 results, spend 25–60 s each reading the full tweet (as reply material).

**Pre-action pause**

```bash
sleep $((RANDOM % 61 + 60))   # 60–120 s — simulate "browse first, then reply"
```

---

### DOM: Scan Replyable Tweets on Current Page

After navigating to the X search results page, scan all tweets with their reply button indices and content.

1. Navigate: `browser-act --session {SESSION} navigate "https://x.com/search?q={KEYWORD_ENCODED}&src=typed_query&f=live"`
   - `{KEYWORD