Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add instructions/replenhq/replen/claude-mdgit clone --depth 1 https://github.com/replenhq/replenWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.04055 | $0.04055 |
| Opus 5 | $0.02027 | $0.02027 |
| Sonnet 5 | $0.00811 | $0.00811 |
| Haiku 4.5 | $0.00405 | $0.00405 |
Grade A, and why
replen CLAUDE.md scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 196 lines — stays where its author put it; the contents beside it link to each section on GitHub.
CLAUDE.md
What it is
Replen is a tool/library discovery service for AI coding workflows. It maps each of a user's repos to its technical capabilities (computer vision, geospatial, market-making, etc.) and watches the OSS ecosystem for things that fit — a library that fills a capability, a new release in a dependency you use, a security advisory in your stack, a standard you implement changing, an upstream going dark. The matching is the product: every candidate is scored against your capabilities, not a generic trending list. The expensive reasoning (per-candidate verdict + writeup) runs in the user's Claude Code / Codex session on their subscription tokens — the server side does only mechanical, no-LLM ingestion + ranking, so there's no per-user LLM bill and source code never leaves the user's machine.
Stack
TypeScript (ESM, Node ≥20) · Next.js 16 + React 19 (dashboard + API routes under src/app/) · Drizzle ORM over SQLite/libSQL (@libsql/client, schema + migrations in src/db/) · node-cron scheduler (src/scheduler/cron.ts, run-once.ts) · Firebase Auth (next-firebase-auth-edge) · Zod validation · cheerio (HTML scraping) · nodemailer (digest email). OpenAI text-embedding-3-small for semantic matching (src/lib/embeddings.ts); a cheap chat model only for catalogue classification (src/catalogue/classify.ts). Ships two sibling npm packages: cli/ (the npx replen onboarding one-liner) and mcp/ (@replen/mcp, a stdio MCP server exposing ~13 tools). A Claude Code skill lives in skills/replen-match/.
Niche / domain
Developer-tooling / AI-agent infrastructure: capability-based OSS discovery, semantic retrieval, and agent-in-the-loop triage. Domain vocabulary that matters: capability facets, facet vs centroid embeddings, modality gate (image/timeseries/text/… cross-modal collision suppression), eligibility filter, discovered vs scouted pools, diversity cap, discovery mode (scouted/discovered/re-checked/prune), Watchtower (no article — the maintained cross-vendor source network the lenses run on: announcement sources + pricing pages + EOL cycles + advisories — ~1,250 sources hosted; starter seeds self-hosted), the watch lenses (libraries / security / stack-releases / spec-changes / upstream-health / pricing), skill tier vs hosted tier, handoff PR, k-anonymity for the cross-user catalogue. Named surfaces (use these terms, no articles on the proper nouns): Brainstem (the matching core — facets, catalogue, learning loop; the sourcing of repos to use as-is / port / cherry-pick / clean-room build), Watchtower (the source network), Atlas (the knowledge graph + webapp; its markdown files are Tiles — never call OUR artifact a 'vault', that's Graphify/Obsidian vocabulary), Leaps, Recall, the Footnote (calm in-session line), the Brief (weekly four-questions email), the Queue.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 2d ago First seen · 196 lines · 4,055 tokens per session scan A 4207ac0e2e08
replen CLAUDE.md is an instructions file published in the GitHub repository replenhq/replen (8 stars, last pushed 22d ago), licensed Apache-2.0. It adds 4,055 tokens to every session, about $0.0203 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other instructions, from other repositories
codevira CLAUDE.md
Instructions for sachinshelke/codevira, covering codevira — persistent project memory, when to call which codevira tool, at the start of every session, before modifying any file and before adopting a pattern, library, or naming convention.
codevira AGENTS.md
Instructions for sachinshelke/codevira, covering codevira-tracked project memory: codevira, locked decisions (donotrevert), active conventions, codevira — persistent project memory and when to call which codevira tool.
codevira GEMINI.md
Instructions for sachinshelke/codevira, covering codevira — persistent project memory, when to call which codevira tool, at the start of every session, before modifying any file and before adopting a pattern, library, or naming convention.
ai-mind-map copilot-instructions.md
Copilot instructions for shdra06/ai-mind-map, covering ai mind map mcp — code memory engine (v1.4.0), 🚀 first thing to do in every conversation, quick lookup: "i need to..." → use this tool, ⚡ session lifecycle (always use these) and 🔍 find code (instead of grep/reading files).
ai-mind-map CLAUDE.md
Claude Code instructions for shdra06/ai-mind-map, covering ai mind map mcp — code memory engine (v1.4.0), 🚀 first thing to do in every conversation, quick lookup: "i need to..." → use this tool, ⚡ session lifecycle (always use these) and 🔍 find code (instead of grep/reading files).
hatch3r CLAUDE.md
Instructions for hatch3r/hatch3r, covering hatch3r — development instructions, architecture, development commands, two-axis pillar framework (2.0.0) and orchestrator self-discipline (bypass protection).