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/dwgx/agentscope/agents-mdgit clone --depth 1 https://github.com/dwgx/AgentScopeWhat 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 | $0.01200 | $0.01200 |
| Opus 5 | $0.00600 | $0.00600 |
| Sonnet 5 | $0.00240 | $0.00240 |
| Haiku 4.5 | $0.00120 | $0.00120 |
Grade A, and why
AgentScope 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.
How it starts
The opening of the file, as written. The whole thing — 106 lines — stays where its author put it; the contents beside it link to each section on GitHub.
AgentScope Repository Working Rules
AgentScope is a Windows-only TypeScript/Electron desktop console for local AI coding agent trace and control. It indexes Codex and Claude Code processes, sessions, transcripts, and evidence. It is not a chat UI and not a Kanban board.
Maintainer And Product Direction
- Speak with the user in Chinese unless they explicitly ask otherwise.
- Be direct, rigorous, and pragmatic. Maintainers expect concrete implementation, screenshots/smoke tests when UI changes, and honest risk reporting.
- The user strongly prefers high-quality UI that feels polished, dense, and controllable. Avoid decorative marketing layouts. Use a serious desktop-console feel with careful spacing, alignment, animation, and keyboard behavior.
- The user dislikes vague claims. Every session/process association must show evidence and confidence.
- Deeper research workflows are useful when explicitly requested, but do not block on delegation if thread limits are reached. Continue locally and document what was verified.
Engineering Defaults
- Use TypeScript, Electron, React, and the existing workspace structure.
- Prefer the existing core library over duplicating logic in the renderer.
- Keep Windows path handling centralized in
packages/core/src/paths.ts. - Keep privileged filesystem/process work in
packages/coreor Electron main. Renderer access goes through preload IPC only. - Do not read or display hidden vendor reasoning. AgentScope only parses plaintext JSONL, SQLite, PID/session maps, path encodings, process metadata, and index relations.
- Default destructive behavior must be blocked, confirmed, backed up, and evidenced.
- Never delete credentials, auth files, settings, plugins, skills, rules, or full global history as a side effect of deleting one session.
Current Commands
Run these before sharing code changes:
npm run audit:repo
npm run lint
npm run typecheck
npm test
npm run i18n:check
npm run package
For release/prebuild verification, run the CI-aligned release check instead:
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 · 106 lines · 1,200 tokens per session scan A 0cd6dd145ff2
AgentScope AGENTS.md is an instructions file published in the GitHub repository dwgx/AgentScope (1 stars, last pushed 1mo ago), licensed MIT. It adds 1,200 tokens to every session, about $0.0060 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
open-codesign CLAUDE.md
Instructions for OpenCoworkAI/open-codesign, covering claude.md — open codesign, what this project is, hard constraints (do not violate), stack & conventions and frontend stack (locked).
opencove AGENTS.md
AGENTS.md instructions for DeadWaveWave/opencove, covering system prompt: opencove project ai developer agent, 1. core directives & golden rules, 2. decision framework (small vs large), a. small change (fast feedback / 小步快反馈) and b. large change (deep thinking / 慎重对齐).
wesight AGENTS.md
Instructions for freestylefly/wesight, covering claude.md, build and development commands, build production bundle (typescript + vite), lint with eslint and run memory extractor tests (node.js built-in test runner).
limboo CLAUDE.md
Instructions for limboo-ai/limboo, covering claude.md, 1. what is limboo?, 2. tech stack (current), 3. project structure and the three electron contexts (critical mental model).
getvect AGENTS.md
Instructions for craigjmidwinter/getvect, covering agents.md, commands (green on a clean checkout), invariants a pr must not break and process.
open-codesign AGENTS.md
Instructions for OpenCoworkAI/open-codesign, covering agents.md - open codesign, what this project is, hard constraints, ai visibility for web work and current architecture direction.