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/robzilla1738/vibe-codr/agents-mdgit clone --depth 1 https://github.com/robzilla1738/vibe-codrWhat 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 | $0.16272 | $0.16272 |
| Opus 5 | $0.08136 | $0.08136 |
| Sonnet 5 | $0.03254 | $0.03254 |
| Haiku 4.5 | $0.01627 | $0.01627 |
Grade A, and why
vibe-codr AGENTS.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.
Copies of this mod
1 near-identical copy found in the catalogue:
- vibe-codr CLAUDE.md — 95% identical, 2 lines differ
How it starts
The opening of the file, as written. The whole thing — 798 lines — stays where its author put it; the contents beside it link to each section on GitHub.
AGENTS.md — vibe-codr
A model-agnostic coding workspace with desktop and terminal interfaces.
TypeScript + Bun monorepo. AGENTS.md is the canonical project
memory; CLAUDE.md exists as the Claude Code bridge and points back here.
Stack & layout
- Runtime: Bun (workspaces + Turbo). Models: Vercel AI SDK v7.
- Hard core/TUI boundary: the engine emits a typed
UIEventstream and acceptsEngineCommands; no UI type leaks into core, so the engine is fully testable headless.
| Package | Owns |
|---|---|
@vibe/shared |
Contracts: UIEvent, Message/Part, ToolDefinition, EngineSnapshot, errors, logger |
@vibe/config |
Zod config schema, file discovery + deep-merge, auth resolution |
@vibe/providers |
ProviderRegistry, resolveModel, CatalogService (models.dev + /v1/models) |
@vibe/tools |
Built-in tools (read/edit/bash/grep/repo_map/git_*/…) + the AI-SDK tool() adapter; the file-write lock is an exclusive-ownership claim registry (createFileLock) so parallel subagents can't clobber one file. Web search is keyless and fans out across DuckDuckGo + Bing (search-engines.ts), then dedupes by canonical URL + quality-ranks the merge (searchcore.ts); TinyFish is an optional booster. Search HTML parsers must keep result/snippet association local to each result row so malformed skipped rows cannot shift later snippets. webfetch extracts PDFs (pdftext.ts, zero-dep) + optional Readability, backed by a cache-through store (fetch-cache.ts). OS sandbox (sandbox.ts, opt-in): a pure Seatbelt(macOS)/bwrap(Linux) policy every command spawn (bash, jobs, and core's exec/verify) routes through — the permission engine stays the policy brain, the sandbox is the kernel backstop |
@vibe/core |
Agent loop (Session.run), Engine, slash commands, checkpoints, context-window tracking, plus three pillars: (1) long-term memory — injected project/global notes (memory.ts), a save_memory write-path (memory-store.ts), and hybrid recall — BM25 (bm25.ts) fused with optional semantic search (embeddings.ts + vector-store.ts over bun:sqlite + semantic-memory.ts) and session recall via RRF (memory-search.ts), behind MemoryService; (2) orchestration — a tree-global AIMD limiter (limiter.ts), the default-ON task-DAG scheduler (orchestrator.ts + orchestration/orchestrator-runner.ts: structured handoffs, read_report, model tiers, executable verify, worktree isolation, ensemble, journal resume), continuation + background spawns (continue_subagent/check_task over a bounded-LRU orchestration/child-registry.ts; detach:true), schema-validated child output (orchestration/structured-output.ts — deterministic local JSON-Schema validation with outputSchema enforced on the inline, worktree, AND ensemble/hard paths: validated JSON or an honest failure, never silently dropped; a continue_subagent that coerced a child to plan mode restores its registry-remembered original mode when continued in execute), and a typed coordination blackboard (blackboard.ts); (2b) build intelligence (build/ — deterministic recon → RepoProfile, run_check parsing, the green-gate, green checkpoints, stub scan, gitops/worktrees, browser verify); (2c) diagnostics — the diagnose() seam behind a composite of the in-process TS fast path and a multi-language lsp/ client (stdio JSON-RPC, lazy per-language spawn, deadline-bounded, advisory-only); (3) MCP (mcp.ts) — stdio + Streamable-HTTP/SSE transports, tools, resources (read_mcp_resource), prompts (get_mcp_prompt) — both network-flagged so permission rules govern them — ${VAR}/${VAR:-default} expansion over connect-time config, OAuth 2.1 (mcp-oauth.ts), and auto-reconnect + tools//resources//prompts/list_changed live re-registration; (4) production — crash handlers + redacted crash log (crash.ts), a keyless update check (update-check.ts) |
@vibe/plugins |
HookBus, slash-command + skill runtimes (SkillRegistry: progressive disclosure, disable-model-invocation / user-invocable frontmatter), PluginHost; declarative shell/HTTP hooks are layered on via core/config-hooks.ts from the config hooks block |
@vibe/tui |
OpenTUI app + headless/REPL renderers, themes, tool icons, spinner |
@vibe/macos-bridge |
Runtime-validated NDJSON stdio host for desktop shells: in-process Engine, same EngineCommand/UIEvent contracts as the TUI. Run bun run macos-bridge; compile with bun run build:macos-bridge → dist/vibecodr-engine-host |
@vibe/cli |
bin/vibecodr entrypoint (argv, config, headless -p vs TUI); the VERSION sentinel (version.ts, stamped at release) and vibe upgrade channel detection (upgrade.ts). Release tooling (binary + npm-bundle builds, version stamping) lives in scripts/release/ |
apps/desktop |
Electron presentation shell, IPC, renderer, Git/diff/terminal panels, Sessions workspace, and desktop packaging. It consumes the engine only through the NDJSON host. |
apps/desktop/mobile |
Expo remote-control client for the authenticated desktop relay; shared engine behavior stays protocol-driven. |
apps/desktop/relay |
Private-LAN/Tailnet or public-WSS handoff relay launched and owned by the desktop main process. |
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 · 798 lines · 16,272 tokens per session scan A 4d87e9ee7335
vibe-codr AGENTS.md is an instructions file published in the GitHub repository robzilla1738/vibe-codr (2 stars, last pushed 1mo ago), licensed MIT. It adds 16,272 tokens to every session, about $0.0814 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
openscience AGENTS.md
AGENTS.md instructions for synthetic-sciences/openscience, covering style guide, avoid let statements, avoid else statements, prefer single word naming and testing.
gini-agent AGENTS.md
AGENTS.md instructions for Open-Curiosity/gini-agent, covering gini agent instructions, shape, adrs, boundaries and branches.
senpi AGENTS.md
Instructions for code-yeongyu/senpi, covering senpi repository guide, mandatory execution protocols — non-negotiable, protocol 1 — explicitly requested modifications, protocol 2 — user-requested pr reviews and structure.
sortie copilot-instructions.md
Instructions for sortie-ai/sortie, covering sortie coding & review standards, 1. layered imports (downward only; violation is critical), 2. concurrency safety, 3. workspace path safety (critical: security boundary) and 4. persistence (sqlite).
copperhead CLAUDE.md
Instructions for copperheadhq/copperhead, covering claude.md, what this repo is, sources of truth, workflow (openspec) and architecture (per spec.md §2).
Xenon CLAUDE.md
Instructions for xianyu-sheng/Xenon, covering claude.md — xenon 项目规范, bug 修复原则, 深度根因分析，而非逐洞补漏, 通用设计，而非特例枚举 and 每次新增 mcp 工具都要加正则 — 不可持续.