Reviewing Agents Md

reviewing-agents-md

Analyze the repository and write (or overwrite) an AGENTS.md file in the project root. The file is a concise, prescriptive orientation guide for AI coding agents. Do not duplicate content from README.md - link there for anything already covered.

Workflow

1. Read README.md, pyproject.toml / package.json / equivalent manifest,

and any existing config files to infer the tech stack and tooling.

1. Identify the environment manager, test runner, linter, and build tool in use.
2. Write AGENTS.md following the structure below.
3. Keep the file under 80 lines total.

Required sections

1. What This Project Does (2-4 sentences)

State the project's purpose and technology stack category. Add a pointer to README.md for full details.

2. Environment & Tooling - CRITICAL

Identify the exact tool that manages the environment and write explicit rules covering:

• The exact command to run the test suite
• How to install/sync dependencies
• How to lint and format
• Which alternatives are explicitly forbidden

Mark this section - CRITICAL so agents cannot miss it.

3. Conventions (non-obvious only)

Cover naming patterns and the canonical source-of-truth files agents respect when adding new parameters or configuration keys.

4. Testing Guidelines

• Where tests live and how they map to source modules
• Whether external I/O needs mocking
• Rules about keeping tests in sync when function signatures change

5. Common Operations

One shell code block with one-liner comments for the 4-6 most frequent tasks (test, lint, build, deploy, docs preview, etc.).

Style rules

• Write in English.
• Use RFC 2119 keywords (MUST, SHOULD, MUST NOT) selectively - reserve them for hard constraints where violation produces broken, incorrect, or insecure output. For routine guidance, use plain imperative language (e.g., "Run tests with pytest", "Format with ruff format"). Overusing RFC 2119 keywords dilutes their signal.
• Omit content already in README.md and link there instead.
• Prefer bullet lists and short paragraphs over prose.
• Do not use emoji, em dashes, or excess bold/italic text.
• Use plain Markdown (no HTML) and only ASCII punctuation.
• When repository facts cannot be determined from the repo contents or user input, do not guess; instead state "Insufficient information".

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: ak1ra-komj
• Source: ak1ra-komj/agents-skills
• License: MIT

Install and usage instructions live in the source repository linked above.