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/alexylon/sofos-code/agents-mdgit clone --depth 1 https://github.com/alexylon/sofos-codeWhat 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 | $0.04921 | $0.04921 |
| Opus 5 | $0.02461 | $0.02461 |
| Sonnet 5 | $0.00984 | $0.00984 |
| Haiku 4.5 | $0.00492 | $0.00492 |
Grade A, and why
sofos-code 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 yesterday.
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 — 477 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Sofos – AI Coding Assistant Project Context
Project Overview
Sofos is a terminal-based AI coding assistant powered by Anthropic's Claude API. It's built in Rust for maximum performance and security. The assistant can read/write files, search code, execute bash commands, and search the web - within the workspace by default, with interactive permission prompts for external paths.
Core Philosophy:
- Security first: Workspace-sandboxed by default, with interactive user-approved access to external paths via three independent scopes (Read, Write, Bash)
- Fast and efficient: Native Rust implementation with optional ultra-fast editing via Morph API
- Developer-friendly: Interactive REPL with session persistence and custom instructions
- Transparent: All tool executions are visible to the user
Architecture
Key Design Decisions
-
Dual Session Storage Format (src/session/history.rs)
api_messages: Anthropic API format for continuing conversationsdisplay_messages: UI-friendly format for showing conversation history- This separation ensures Claude sees proper API format while users see original UI
-
Tool Calling Pattern (src/repl/mod.rs)
- Assistant returns content blocks (text + tool_use)
- REPL executes tools and collects results
- Results sent back as user message with tool_result blocks
- Loop-based handling allows Claude to use multiple tools in sequence iteratively
-
Two-Level Instructions (src/session/history.rs)
AGENTS.md: Project-level, version controlled.sofos/instructions.md: Personal, gitignored- Both appended to system prompt at startup
-
Sandboxing Strategy (src/tools/filesystem.rs, src/tools/bashexec.rs)
- All paths validated before operations
- Parent directory traversal blocked (
..) - Absolute paths rejected
- Symlinks checked to prevent escape
- Bash commands filtered through blocklist
Code Organization
Directory Structure
src/
├── main.rs # Entry point
├── cli.rs # CLI argument parsing
├── error.rs # Error types
├── error_ext.rs # Error extensions
├── config.rs # Configuration (SofosConfig, ModelConfig)
│
├── api/ # API clients
│ ├── anthropic.rs # Claude API client
│ ├── openai.rs # OpenAI API client
│ ├── morph.rs # Morph Apply API client
│ ├── types.rs # Message types and serialization
│ └── utils.rs # API utilities
│
├── mcp/ # MCP (Model Context Protocol) integration
│ ├── mod.rs # MCP module exports
│ ├── config.rs # MCP server configuration loading
│ ├── protocol.rs # MCP protocol types (JSON-RPC, tools)
│ ├── client.rs # MCP client implementations (stdio, HTTP)
│ └── manager.rs # MCP server connection management
│
├── repl/ # REPL components
│ ├── mod.rs # Main REPL loop and Repl struct
│ ├── clipboard_edit_mode.rs # Custom EditMode wrapping Emacs to intercept Ctrl+V
│ ├── conversation.rs # Message history management
│ ├── prompt.rs # Prompt rendering
│ ├── request_builder.rs # API request construction
│ └── response_handler.rs # Response processing
│
├── session/ # Session management
│ ├── history.rs # Session persistence + custom instructions
│ ├── state.rs # Runtime session state
│ └── selector.rs # Session selection TUI
│
├── clipboard.rs # Clipboard image paste (Ctrl+V) with numbered markers
│
├── tools/ # Tool implementations
│ ├── filesystem.rs # File operations (read, write, list, etc)
│ ├── bashexec.rs # Sandboxed bash execution
│ ├── codesearch.rs # Ripgrep integration
│ ├── image.rs # Image handling (local paths, URLs)
│ ├── plan.rs # Task-plan validation and terminal rendering
│ ├── permissions.rs # 3-tier command permission system
│ ├── tool_name.rs # Type-safe tool name enum
│ ├── types.rs # Tool definitions for API
│ └── utils.rs # Tool utilities (confirmations, HTML-to-text)
│
├── ui/ # UI components
│ ├── mod.rs # Main UI utilities and display logic
│ ├── syntax.rs # Markdown/code syntax highlighting
│ └── diff.rs # Contextual diff generation and display
│
└── commands/ # Built-in commands
└── builtin.rs # Command implementations
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.
- yesterday First seen · 477 lines · 4,921 tokens per session scan A 5fca77092b97
sofos-code AGENTS.md is an instructions file published in the GitHub repository alexylon/sofos-code (10 stars, last pushed 1mo ago), licensed Apache-2.0. It adds 4,921 tokens to every session, about $0.0246 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
stitchfy CLAUDE.md
Instructions for devifyllc/stitchfy, covering claude.md, commands, architecture, pipeline flow — blueprint (shared) and pipeline flow — template generator (build:site).
patchloom AGENTS.md
Instructions for patchloom/patchloom, covering agents.md, project overview, dev commands, git hygiene and release prs (release-please).
Perigon.CLI copilot-instructions.md
Instructions for AterDev/Perigon.CLI, covering github copilot instructions, general guidelines, 技术栈, 项目结构与分层 and 代码风格约定.
ducklab AGENTS.md
Instructions for jrullan/ducklab, covering working in this repository as an agent and path 1: conventional (edit the tree directly).
sdlc-studio CLAUDE.md
Instructions for DarrenBenson/sdlc-studio, covering claude.md and claude code specifics.
spine CLAUDE.md
Instructions for synaptixs/spine, covering spine — working on this repo, layout — where things live, invariants — break these and things get subtly wrong, gotchas that have bitten and before pushing.