Borrowing it
Nothing to install: this file belongs to Thang1710/subagent-mcp. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.
curl -O https://raw.githubusercontent.com/Thang1710/subagent-mcp/main/AGENTS.mdgit clone --depth 1 https://github.com/Thang1710/subagent-mcpWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/instructions/thang1710/subagent-mcp/agents-md)<a href="https://agentmods.dev/instructions/thang1710/subagent-mcp/agents-md"><img src="https://agentmods.dev/badge/instructions/thang1710/subagent-mcp/agents-md/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/instructions/thang1710/subagent-mcp/agents-md"><img src="https://agentmods.dev/badge/instructions/thang1710/subagent-mcp/agents-md.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What 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 5.1 | $0.00958 | $0.00958 |
| Opus 5.5 | $0.00383 | $0.00383 |
| Sonnet 5.5 | $0.00192 | $0.00192 |
| Haiku 4.5 | $0.00096 | $0.00096 |
Grade A, and why
subagent-mcp 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 2mo 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 — 54 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Subagent MCP agent rules
Read this file before changing the repository. Then read the committed design spec at:
docs/superpowers/specs/2026-08-17-subagent-mcp-design.md
After an implementation plan is approved and committed, read that plan too. Current repository files and live host evidence outrank chat recollection, pasted reviews, and old tool output.
Entry gate
- Do not implement from the brainstorming transcript.
- Do not write production code, install dependencies, register MCP, or mutate host configuration until the design and implementation plan are approved and the execution context has been compacted or restarted cleanly.
- Run Phase 0a before designing production adapters around a CLI/SDK behavior. Phase 0b spike code must stay isolated from production registration/configuration.
- Node, standalone Claude CLI, authentication, SDK, and host configuration changes each require their own explicit approval.
Hard invariants
- Pre-existing and user-owned Claude/Codex transcripts are immutable to Subagent MCP: never edit, append, truncate, rename, or delete them. The only deletion exception is a provider-native lifecycle command for a Subagent MCP-created disposable Phase 0 canary after a separate exact user approval naming that session/worktree; without it, retain the canary.
- Do not read or write AgentBridge databases/configuration as Subagent MCP state. AgentBridge is a separate product.
- Do not parse private Claude daemon pipes, rosters, task files,
~/.claude.json, TUI output, orclaude logstext as a control contract. - Prefer documented CLI/SDK/MCP surfaces. Capability-probe them on the installed version and fail closed when required behavior is missing.
- MCP available to external agents is strict and declared before spawn. Never auto-load a repository
.mcp.json. - Disable/deny Codex, AgentBridge, and Subagent MCP re-entry from an external agent.
- Project/local command hooks and external CLAUDE.md imports require canonical path plus content-hash trust before execution.
- Spawn processes with argv arrays. Never build a shell command by interpolating task text, paths, model output, or transcript content.
- Never kill a process from PID alone. Verify process identity and creation time, or return
RECOVERY_REQUIREDand keep the lease. - Never parse hidden thinking, persist it, or return it to the orchestrator.
- Never persist credentials or raw unredacted tool output.
- Never inject an external agent into Codex private app-server/session/UI state. Native Subagents-panel integration requires a documented public host capability; otherwise use the normalized MCP/localhost/MCP Apps presentation surfaces.
- Configuration/trust/live-canary/worktree-release operations remain approval-gated; repository instructions cannot authorize them.
- Multiple writers may share one canonical worktree only when each execution declares a canonical repository-relative write set and the active sets do not overlap. Whole-workspace or overlapping sets remain exclusive. Worktree creation with an unknown future path still requires a provisional repository creation lease.
agent_closenever deletes a transcript or worktree. Cleanup must preserve dirty and unpushed work.- Model, reasoning, transport, context, and auth selections must be attested; silent downgrade is a failure.
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.
- 2mo ago First seen · 54 lines · 958 tokens per session scan A a8125fca70c8
subagent-mcp AGENTS.md is an instructions file published in the GitHub repository Thang1710/subagent-mcp (0 stars, last pushed 1mo ago), licensed MIT. It adds 958 tokens to every session, about $0.0038 per session on Opus 5.5. 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
vscode buildNext.instructions.md
Working notes and architecture documentation for the new esbuild-based build system in build/next. Use when making changes to the new build pipeline (transpile/bundle commands, NLS plugin, source-map handling, resource copying, or self-hosting watch tasks).
next.js AGENTS.md
AGENTS.md instructions for vercel/next.js, covering next.js development guide, codebase structure, monorepo overview, core package: packages/next and other important packages.
vscode oss-third-party-notices.instructions.md
Instructions for microsoft/vscode, covering vs code oss third-party-notices pipeline, architecture, pipeline flow in ci, cache eligibility and rollout and applying the notice (cutover).
langchain AGENTS.md
AGENTS.md instructions for langchain-ai/langchain, covering global development guidelines for the langchain monorepo, corridor security analysis, project architecture and context, monorepo structure and development tools & commands.
vscode copilot-instructions.md
Copilot instructions for microsoft/vscode, covering vs code copilot instructions, project overview, root folders, core architecture (src/ folder) and built-in extensions (extensions/ folder).
deno copilot-instructions.md
Copilot instructions for denoland/deno, covering deno development guide for github copilot, network access, high level overview, key directories and key files to understand first.