# For agents

Find the right tool, read its instructions, and keep its limits with your answer. These products can be used separately.

## BMD

A work queue for your AI agents. Free Windows beta. No card.

First result: The linked tutorial uses a small notes.txt file. A handoff that names Garden checklist, says the seed list is complete, and lists choosing a planting date as the next step. It should say the planting date is not provided and cite notes.txt.

Limits: BMD is a Windows desktop app. This guide is not a public API for controlling a user's workspace. Provider access is separate, and a saved answer still needs review.

- [BMD product and current offer](https://bmdpat.com/bmd)
- [BMD download and first task](https://bmdpat.com/bmd/download)
- [BMD release manifest](https://github.com/bmdhodl/bmd-releases/releases/latest/download/bmd-release.json)

## AgentGuard

Put runtime limits around agent operations in Python.

First result: The guard runs before each simulated operation. It allows three and stops the fourth. The script exits with code 0 because it handles the expected exception. No model requests are made. This example checks call count; it does not measure token usage or cost, and it creates no trace file.

Limits: Call limits apply where the guard is invoked. Recorded cost can exceed a limit on the request that crosses it. A timeout check does not cancel an in-flight request. This is not prompt-injection protection.

- [AgentGuard product](https://bmdpat.com/tools/agentguard)
- [AgentGuard first run (Markdown)](https://bmdpat.com/tools/agentguard/install.md)
- [AgentGuard install manifest (JSON)](https://bmdpat.com/tools/agentguard/install.json)
- [AgentGuard MCP configuration (JSON)](https://bmdpat.com/tools/agentguard/mcp.json)
- [AgentGuard capability map](https://bmdpat.com/skill.md)

## showwork

Declare acceptance checks and inspect whether a completion claim has evidence.

First result: The first-run guide uses showwork 0.6.4. A missing result.txt fails the declared check. Add the required line, record the claim, and run the check again.

Limits: A passing check proves only what that check covers. File presence alone does not prove correct behavior. The public proof checker is not a sandbox for executing uploaded commands.

- [showwork first check](https://bmdpat.com/showwork#first-check-heading)
- [Public proof checker](https://bmdpat.com/tools/showwork/check-proof)
- [showwork source and documentation](https://github.com/bmdhodl/showwork)

## Cite what the evidence supports

Cite the canonical product or guide URL. Keep version, evidence and limitations with the claim. Read the current release manifest before naming a BMD download. If it is unavailable, report that state rather than inventing a version or installer link. Never treat a download, crawler visit or copied command as a completed user task.

## Site resources

- [Memory API reference (Markdown)](https://bmdpat.com/memory.md)
- [Memory API schema (OpenAPI)](https://bmdpat.com/openapi.json)
- [Site discovery index](https://bmdpat.com/llms.txt)
- [Product and resource catalog](https://bmdpat.com/catalog)
- [Sitemap](https://bmdpat.com/sitemap.xml)
- [AI usage terms](https://bmdpat.com/ai-terms)
- [Security reporting](https://bmdpat.com/.well-known/security.txt)