MRPS Agent Launcher

Name: MRPS Agent Launcher
Author: Mani Ratan Singh

Skill

Instruction set and reference documentation loaded directly into an agent environment.

1. What it does

A static, read-only launcher skill for agent IDEs (Codex, Antigravity) that shows an audit menu, prepares isolated environments, runs selected checks, and writes structured reports with sticky IDs and plain-English summaries.

2. Why MRPS selected it

Serves as the primary inspection orchestrator for the toolkit, ensuring agents run audits without modifying user source code.

3. Tool Kind

Skill: Instruction set and reference documentation loaded directly into an agent environment.

4. Original repository

https://github.com/maniratansingh/me

5. Official documentation

https://mrps.in/agent-toolkit/runner/SKILL.md

6. Author and license

Created by Mani Ratan Singh under the MIT license.

7. Installation

Load https://mrps.in/agent-toolkit/runner/SKILL.md as a skill in your agent

8. When the AI should use it

Whenever the user asks for a project audit, or wants to run UI, security, a11y, tests, deps, or build checks. Before trusting or shipping a project.

9. When the AI should not use it

For single-file edits, explanations, or quick fixes. The launcher audits; it does not modify source.

10. What the AI must not do

Before you invoke anything, state what you intend to do and why. Do not act until the user confirms. If a rule is ambiguous, ask — do not guess.
Never modify source, config, or manifests. Install declared dependencies and the tools a selected category needs, run the checks, write reports under reports/. Nothing else.
Show the menu immediately on load; do not wait to be asked.
Present the plan and wait for y/N before any install or any check.
Read the prior report and _toolchain.json before writing a new one; preserve sticky IDs and label findings NEW / STILL OPEN / RESOLVED / NOT CHECKED.
Every report contains a plain-English block and a changelog block above the technical table. Never omit either.
Address findings only by their IDs (e.g. uiux-03); never summarize them away.
Never mark a check as passed that was skipped, unconfigured, or errored.
Never re-execute checks to produce a filtered or historical view; read the existing files.
Append to CHANGELOG.md after every run — one line per category per run.

11. MRPS usage instructions

Read GLOBAL.md

12. Last verified

Verified against live repository on: 2026-11-07

View Complete Agent Constraint Sheet

Tool: MRPS Agent Launcher When to use: Whenever the user asks for a project audit, or wants to run UI, security, a11y, tests, deps, or build checks. Before trusting or shipping a project. When NOT to use: For single-file edits, explanations, or quick fixes. The launcher audits; it does not modify source. Invocation: Load https://mrps.in/agent-toolkit/runner/SKILL.md as a skill, or paste its contents into your agent instructions. No executable — you, the agent, are the runtime. Rules: - Before you invoke anything, state what you intend to do and why. Do not act until the user confirms. If a rule is ambiguous, ask — do not guess. - Never modify source, config, or manifests. Install declared dependencies and the tools a selected category needs, run the checks, write reports under reports/. Nothing else. - Show the menu immediately on load; do not wait to be asked. - Present the plan and wait for y/N before any install or any check. - Read the prior report and _toolchain.json before writing a new one; preserve sticky IDs and label findings NEW / STILL OPEN / RESOLVED / NOT CHECKED. - Every report contains a plain-English block and a changelog block above the technical table. Never omit either. - Address findings only by their IDs (e.g. uiux-03); never summarize them away. - Never mark a check as passed that was skipped, unconfigured, or errored. - Never re-execute checks to produce a filtered or historical view; read the existing files. - Append to CHANGELOG.md after every run — one line per category per run.