I ran a Claude Code agent on growth for a week behind a one-word approval gate — here's what worked
intensityflow · reddit · 2026-10-02
The author handed growth for a free Chrome extension to a Claude Code agent: a nightly scheduled task loads a markdown playbook covering metrics, comment replies, target threads, and a Notion report.
The guardrails were the interesting part:
- One-word approval: nothing posts until the author replies "go" (or "go A1 A3" for a subset); silence means nothing goes out
- Hard limits: no account creation, passwords, captchas, vote begging, or impersonation; respect subreddit self-promo rules
- Verify which account is signed in before touching pages (learned the hard way)
- Fenced self-edits: fixes go into its own playbook; decision- or approval-changing edits wait for humans
- All state in plain files: post metrics at 24h/72h/7d, backlog, scoreboard
Wins: caught a near-identical named competitor before launch, fixed a stale store listing, built and tested the next version (122 checks, 55 in a real browser).
Failures: Google's periodic password prompt stalled a release 5 days; first posts got single-digit upvotes; the agent suggested texting friends (vetoed).
5-day result: 4 weekly users, $0 spent, next version in review.
Related event: Developer Lets Claude Code Autonomously Run Plugin for a Week(2 posts)→
More from coding & agent
- Gefei Launches SEO Agent: One Sentence Triggers Keyword, Domain & Site Analysis — gefei55 · 2026-10-02
- Voice-agent lesson: look up orders early, don't cancel mid-conversation — HaktanSuren · 2026-10-02
- Someone launched an autonomous X account that independently researches and questions the AI field — irinarish · 2026-10-02
- User Uses GPT-6.1 to Reconcile Dozens of AI Subscriptions on Just 2% of Weekly Limit — EXM7777 · 2026-10-02
- Parse XML Like JSON in Python with xmltodict — KhuyenTran16 · 2026-10-02
- Designer Builds AI Model 'Tarot Deck' Site, Codex Agent Auto-Adds New Models — ColleenMBrady · 2026-10-02