# X / Twitter launch thread β€” paste-ready > Post Tuesday 8:30am PT, same minute as the Show HN. Pin the first tweet. ## Tweet 1 β€” pinned, attach the 60-sec video ``` I open-sourced doc-wiki today. One command turns your codebase into a wiki your coding agent reads β€” ER diagrams from your ORM models, a cross-service map, cited answers over code + Jira + Confluence + DB schemas. I benchmarked it and published the null. Apache 2.0 forever. 🧡 [VIDEO β€” 60-sec demo, from media/demo.mp4] ``` **Counts:** 279/280 (X weighted count; the 🧡 emoji counts as 2). The thread emoji at the end signals "more below" to the algorithm. ## Tweet 2 β€” the why ``` Most "Claude Code makes me 10Γ— more productive" stories are on clean OSS code. The codebase you actually work in is 8 years old, glued to Jira tickets nobody reads, a DB schema that drifted from the ORM 3 refactors ago. That's the case I built doc-wiki for. ``` ## Tweet 3 β€” the pattern ``` The pattern: Karpathy's LLM Wiki (https://gist.github.com/karpathy/442a6bf555914893e9891c11519de94f) pointed at enterprise code instead of personal notes. Works on small repos and root-of-microservices layouts too. A maintained, compounding artifact the agent reads instead of re-deriving everything each turn. ``` **Counts:** 268/280 (URL counts as 23 via t.co). ## Tweet 4 β€” the 8 commands ``` Eight slash commands cover the lifecycle: /doc-wiki:init β†’ scaffold + onboard /doc-wiki:atlas β†’ document the whole codebase /doc-wiki:ingest β†’ add a source; --refresh /doc-wiki:query β†’ cited answers; --promote /doc-wiki:lint, :edit, :unarchive, :stats ``` **Counts:** 251/280. ## Tweet 5 β€” the benchmark ``` The honest part: I benchmarked whether the wiki lifts fully-autonomous ticket-fix rate. 92 ticket pairs, SWE-bench-style, container-isolated. It doesn't: baseline 57/92, wiki 54/92. Null published in full β€” re-run it yourself and argue with the data. github.com/narailabs/doc-wiki/blob/main/benchmark/RESULTS.md ``` **Counts:** 275/280 (URL counts as 23 via t.co). ## Tweet 5b β€” why use it then (post immediately after 5, same thread) ``` So why use it? The benchmark measures an agent alone with one OSS repo. The wiki earns its keep when you + the agent work ecosystem-heavy code: scope from the cross-service map, point at the right pages, catch wrong turns at diff two. The artifact is the product. ``` **Counts:** 263/280. ## Tweet 6 β€” the license ``` Apache 2.0. Forever. No relicensing, no rug-pull, no "doc-wiki Pro" with the good parts behind a paywall. Explicit no-relicense commitment in docs/governance.md. The Terraformβ†’OpenTofu and Redisβ†’Valkey lessons cost the ecosystem too much already. ``` ## Tweet 7 β€” call to action ``` Install: claude plugin install narailabs/doc-wiki 5-min walkthrough: github.com/narailabs/doc-wiki/blob/main/docs/getting-started.md Manifesto (why this exists, what "LLM Wiki for enterprise code" means, the standard-track argument): github.com/narailabs/doc-wiki/blob/main/docs/manifesto.md Built it solo. Looking for users + feedback + arguments. ``` --- ## Tagging strategy **Tweet 1 (pinned)** β€” no tags. Tags in a pinned tweet hurt reach because the algorithm reads them as a thread-of-replies signal. **Tweet 6 or 7** β€” tag `@bcherny @alexalbert__ @ClaudeDevs` *once*, naturally in the body (not as an `.@` at the start). Don't tag Karpathy, swyx, or Simon Willison in-thread β€” DM them separately. Tagging celebrities in-thread reads as spam. **Subsequent threads (T+1 to T+7)** β€” use the hashtag `#claudecode` sparingly. It's a real tag but oversaturated. ## Engagement plan - Reply to every reply for 24h. - For replies that are questions: thank β†’ answer crisply β†’ link to the right doc. - For replies that are criticism: thank β†’ engage genuinely β†’ don't get defensive. The audience watches. - For replies that are spam ("nice work bro, check out my product"): mute, don't engage. - For replies that are RTs from named accounts (Boris, swyx, Karpathy, Theo): quote-reply with a single thoughtful tweet that adds new info. Don't gush. ## Post-thread cadence (T+8 onward) One tweet every 2 days. Three rotation types: 1. **User story** β€” "X engineer just used doc-wiki on Y codebase, here's what changed." 2. **Feature drop** β€” "Today shipped Z; the why." 3. **Karpathy LLM Wiki community-implementation roundup** β€” quote-tweet someone in the wider LLM Wiki ecosystem with a thoughtful adjacency to doc-wiki. Don't tweet-bait. Don't ratio-bait. Don't reply-guy on @bcherny tweets unless you have something genuinely useful to add.