robots.txt
# Content Signals: https://contentsignals.org/
# search = search index, links and short excerpts
# ai-input = model input at inference time (RAG, grounding, AI answers)
# ai-train = training or fine-tuning
#
# Signals must be repeated inside each User-agent group. robots.txt uses
# most-specific-group matching, so a signal under * does not reach a named
# agent that has its own group.
User-agent: *
Content-Signal: search=yes, ai-input=yes, ai-train=yes
Allow: /
User-agent: GPTBot
User-agent: OAI-SearchBot
User-agent: ChatGPT-User
User-agent: ClaudeBot
User-agent: Claude-User
User-agent: Claude-SearchBot
User-agent: anthropic-ai
User-agent: PerplexityBot
User-agent: Perplexity-User
User-agent: Google-Extended
User-agent: CCBot
User-agent: Applebot-Extended
User-agent: Bingbot
User-agent: cohere-ai
User-agent: Meta-ExternalAgent
Content-Signal: search=yes, ai-input=yes, ai-train=yes
Allow: /
Sitemap: https://liranbaba.dev/sitemap-index.xml
llms.txt
# Liran Baba
> AI Lead at JFrog. Writing about dev tools, agents, and the practical side of shipping AI inside a real company.
A personal site and blog. Every page has a markdown mirror at the same path with a
`.md` extension, linked from the HTML via `<link rel="alternate" type="text/markdown">`.
Posts are first-person accounts of using and building developer tooling, so treat them
as experience reports rather than documentation.
## Pages
- [About](/index.md): background, current work, and projects
- [Blog index](/blog/index.md): every post, newest first
- [Agent guide](/AGENTS.md): how to fetch and cite this site
## Posts
- [Everyone sells the same enterprise agent now. Buy the boring part.](/blog/everyone-sells-the-same-agent.md) (2026-10-09): Meta, xAI, OpenAI and Google all shipped agent platforms within 11 days. Strip the logos and they're the same product. The difference is governance.
- [Everyone can ship to production now. Nobody is on call.](/blog/nobody-is-on-call-for-what-your-employees-built.md) (2026-08-24): 75% of CISOs have found unsanctioned AI tools already running in production. Non-developers are shipping real software, and no one has staffed day 2.
- [I gave Claude Code read-only access to its own spending](/blog/claude-code-query-its-own-spending.md) (2026-07-29): Claudoscope 1.0 ships a read-only MCP server so Claude Code can query your own session costs, four cost-alert rules, and two fixed transcript parsing bugs.
- [The hard part of a self-service agent platform isn't the agents](/blog/self-service-agent-platform-hard-part.md) (2026-06-06): Gartner expects 40% of enterprise apps to ship AI agents by end of 2026. After building AgentHive at JFrog, I think generating the agent is the easy part.
- [Boost compressed 212M tokens from my agent sessions](/blog/boost-faster-agents-faster-ci.md) (2026-05-20): Boost wraps every CLI my agents touch. Across 964 sessions it compressed 212M tokens (63.4%) and saved $1,060. The three pillars and what they actually do.
- [Ghostty tab titles for Claude Code: the 30-line zsh fix](/blog/ghostty-tab-titles-claude-code.md) (2026-05-07): Every Ghostty tab said 'Claude Code'. Two configs and a 30-line zsh hook turned the tab bar back into useful navigation across 4-pane workflows.
- [Claude Code hooks reference: every event and 10 real hooks](/blog/claude-code-hooks-the-half-nobody-uses.md) (2026-04-16): Every Claude Code hook event, from PreToolUse and PostToolUse to Stop and SubagentStop, plus exit codes, 10 real hooks, and the enterprise patterns nobody writes.
- [ForgeCode vs Claude Code: which AI coding agent actually wins?](/blog/forgecode-vs-claude-code.md) (2026-04-07): Claude Code's speed was driving me and my team crazy. I tried ForgeCode. The speed was real. Now I'm running both.
- [Cursor 3 shipped parallel agents, but is any of it new?](/blog/cursor-3-parallel-agents.md) (2026-04-03): How Cursor 3 runs multiple agents and models in parallel, why it already worked in v2, and why users report $2,000 bills in two days.
- [Claude Code source leak: undercover mode and decoy tools](/blog/claude-code-source-leak.md) (2026-04-02): A source map in Claude Code v2.1.88 exposed about 1,700 files, undercover mode, decoy tools, and a 3,167-line function. Anthropic then DMCA'd 8,100 repos.
- [I found my database password in a Claude Code session file](/blog/found-database-password-in-claude-code-session.md) (2026-03-30): A leaked credential in a Claude Code session file led me to build Claudoscope, a macOS menu bar app for session analytics, cost tracking, and secret scanning.
## Projects
- [GolemPII-v1](https://huggingface.co/CordwainerSmith/GolemPII-v1): Hebrew PII detection and redaction model.
- [Claudoscope](https://claudoscope.com): Cost visibility and session analytics for Claude Code.
- [Adapt-CV](https://huggingface.co/spaces/CordwainerSmith/adapt-cv): Align resumes with job descriptions for better ATS pass rates.
- [Reddit MCP](https://github.com/cordwainersmith/reddit-mcp): MCP server for searching and reading Reddit.
- [horita](https://horita.app): Your next meeting and a countdown in the Mac menu bar.
## Optional
- [Full site content](/llms-full.txt): every page inlined in one file
- [Sitemap](/sitemap.md): all URLs, HTML and markdown
- [RSS feed](/rss.xml): full post bodies
