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/provos/ironcurtain/copilot-instructionsgit clone --depth 1 https://github.com/provos/ironcurtainWrote 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/provos/ironcurtain/copilot-instructions)<a href="https://agentmods.dev/instructions/provos/ironcurtain/copilot-instructions"><img src="https://agentmods.dev/badge/instructions/provos/ironcurtain/copilot-instructions.svg" alt="Measured on agentmods" 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 | $0.00850 | $0.00850 |
| Opus 5 | $0.00425 | $0.00425 |
| Sonnet 5 | $0.00170 | $0.00170 |
| Haiku 4.5 | $0.00085 | $0.00085 |
Grade A, and why
ironcurtain copilot-instructions.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 6d 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 — 48 lines — stays where its author put it; the contents beside it link to each section on GitHub.
IronCurtain Copilot Review Instructions
IronCurtain is a secure agent runtime that mediates between an AI agent and MCP servers. Every tool call passes through a trusted process that evaluates it against policy rules before forwarding. The policy engine and MCP proxy are the security boundary.
Platform Support
IronCurtain runs on macOS and Linux. Windows is only supported under WSL2 — native Windows is not a target platform.
Security Invariants
- Every tool call MUST pass through
policyEngine.evaluate()before forwarding to a real MCP server. Any code path that skips evaluation is a security bypass. - Trusted servers are an intentional exception to annotation requirements. Servers in the
trustedServersset (currently only the memory MCP server) bypass structural invariants and compiled rules via an early return inevaluate(). This is by design: trusted server tools have no filesystem paths or URLs, so structural invariants (protected paths, sandbox containment, domain gate) are not applicable and would require fake annotations. Trust is verified by matching both the server name AND the expected entrypoint path (MEMORY_SERVER_ENTRY), not name alone. The normal flow (annotation lookup → evaluate → forward → audit) still runs; onlyprepareToolArgsis skipped when no annotation exists. - Path containment checks MUST use
resolveRealPath()(symlink resolution) before comparison. Rawpath.resolve()is a symlink-escape vulnerability. isWithinDirectory()appends a trailing/beforestartsWith(). Without it,/tmp/sandbox-evilmatches/tmp/sandbox.- Multi-role evaluation uses most-restrictive-wins (deny > escalate > allow). Least-restrictive is a privilege escalation.
- Unknown tools (without annotations) MUST be denied, unless the server is in the
trustedServersset. - The auto-approver can only return
approveorescalate, neverdeny. Error paths must returnescalate(fail-open to human). SERVER_CREDENTIALSmust be deleted fromprocess.envimmediately after parsing to prevent child process leakage.- Audit logging must occur for every tool call outcome. No code path should skip it.
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.
- 6d ago First seen · 48 lines · 850 tokens per session scan A 717b57c86929
ironcurtain copilot-instructions.md is an instructions file published in the GitHub repository provos/ironcurtain (600 stars, last pushed yesterday), licensed Apache-2.0. It adds 850 tokens to every session, about $0.0042 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 2026-08-30.
Other instructions, from other repositories
agentos CLAUDE.md
Claude Code instructions for rivet-dev/agentos, covering agentos, boundaries, security model, sqlite schema ownership and runtime and registry.
agent-sandbox AGENTS.md
AGENTS.md instructions for kubernetes-sigs/agent-sandbox, covering agents.md, project summary, repository layout, agent skills and build, test, lint.
agent-sandbox copilot-instructions.md
Copilot instructions for kubernetes-sigs/agent-sandbox: Project Context & Architecture: Refer to AGENTS.md for full project background, module layout, toolchain versions, and core conventions.
mcp-server-starrocks CLAUDE.md
Claude Code instructions for StarRocks/mcp-server-starrocks, covering claude.md, project overview, development commands, run the server directly for testing and run with test mode to verify table overview functionality.
after-effects-mcp AGENTS.md
AGENTS.md instructions for JUNKDOGE-JOE/after-effects-mcp, covering repository development and delivery rules, 0.0 read the current architecture direction first, 0. user authorization is the scope boundary, 1. measure outcomes, not activity and 2. prioritize by dependency and user value.
Symbiont AGENTS.md
AGENTS.md instructions for ThirdKeyAI/Symbiont, covering symbiont — agent instructions, project structure, build and test, code style and commit guidelines.