Get started

Running in about five minutes.

Pick your agent, copy four commands, then send one bounded request. There is a single runtime and no profile to choose — every host gets the same 26 top-level skills.

Python 3.11+
git
5 supported agents

Step 1

Install it for your agent.

The options below are read from the installer's own tables, so every combination shown is one the CLI accepts. Preview first — step 3 writes nothing.

Which agent do you use?

Where should it be installed?

Available in every repository you open. Start here.

Files land in: ~/.agents/skills
Agent profiles: Native main, analysis, task and review profiles

01Get the repository

git clone https://github.com/machenjie/rd-skills.git
cd rd-skills

02Install the Python package (3.11+)

python3 --version
python3 -m pip install .

03Preview — writes nothing

python3 scripts/quickstart.py --agent codex --scope user --dry-run

04Build, install, run doctor

python3 scripts/quickstart.py --agent codex --scope user

05Re-check any time

python3 installers/doctor.py --agent codex --scope user

Step 2

Send one bounded request.

Open a small test repository in your agent and paste this. The five lines are the whole contract — they tell rd-skills what done means and when to stop.

Your first task

/engineering-control-plane

Goal: Add an empty-string check to src/example.py without changing its public API.
Acceptance: Empty input returns the existing validation error; valid input is unchanged.
Allowed scope: src/example.py and its existing test file only.
Verify: Run python3 -m unittest tests.test_example.
Stop if: Ownership, the public contract, or the verification command differs.

Goal: What should be different when this is done. One sentence.
Acceptance: How anyone could observe that it worked, without reading the diff.
Allowed scope: The files it may touch. Everything else is off limits.
Verify: The exact command that proves it. It runs after the final edit.
Stop if: The condition that should make it come back and ask instead of guessing.

Some hosts have no native slash UI. Typing /engineering-control-plane into the request text still expresses routing intent — it just does not mean the host implements slash commands.

Step 3

Know what a good result looks like.

If you get a diff and nothing else, something is wrong. A bounded implementation should produce all five of these.

one primary professional skill takes ownership
a task agent implements inside the scope you gave
validation runs after the final edit, not before
a separate review agent reads the actual diff
the handoff lists changed files, results, unverified scope and residual risk

A healthy doctor run

Doctor confirms 26 installed skills, the manifest, current source bindings and the agent profile expectation for your host. It proves the files are correct on disk — not that the host has loaded them.

If setup fails

Read the conflict before reaching for --force. An old recommended, full or dev manifest means migration: run upgrade rather than uninstalling first.

Troubleshooting and recovery →

Three request patterns Browse the 25 professionals Repository

Reference

Quickstart

The complete document from the repository, including project scope, the OpenAI API zip path and the exact expected outcome.

docs/QUICKSTART.md

This path starts at a fresh checkout and ends with a first bounded engineering request. Run commands from the repository root.

Prepare

rd-skills requires Python 3.11 or newer. Install its declared dependencies:

python3 --version
python3 -m pip install .

Do not install src/ or source registries. Quickstart builds the one Runtime into dist/ before using the installer. There is no Profile choice: every Host receives the same 26 top-level Skills, consisting of 1 Control and 25 Professional Skills. Foundation and Domain knowledge remains behind targeted JIT selectors.

Choose A Host And Scope

Supported hosts are codex, claude, copilot, cline, and openai-api. Codex supports project, user, and admin; Claude, Copilot, and Cline support project and user. OpenAI API produces zip files and has no installation scope. Codex, Claude, and Copilot receive four native Agent Profile files; Cline and OpenAI API receive standard Skills only.

Runtime and Agent Profile are different concepts. Runtime is the fixed Skill surface and JIT knowledge path. The four Agent Profiles are the static main, analysis, task, and review execution roles; users do not select among them as an installation profile.

Use user for a personal default. Use project to keep installation inside a specific checkout; replace /absolute/path/to/project below with the real project root. Use Codex admin only with explicit administrative approval; see Installation.

Preview Before Writing

Preview a user installation:

python3 scripts/quickstart.py --agent codex --scope user --dry-run

Preview a project installation:

python3 scripts/quickstart.py --agent claude --scope project --target /absolute/path/to/project --dry-run

The preview prints the build, install, and doctor plan but does not execute it.

Install And Run Doctor

Run the same plan without --dry-run:

python3 scripts/quickstart.py --agent codex --scope user

For project scope:

python3 scripts/quickstart.py --agent claude --scope project --target /absolute/path/to/project

Quickstart builds, installs, and runs doctor for Codex, Claude, Copilot, and Cline. Repeat doctor independently when diagnosing a changed installation:

python3 installers/doctor.py --agent codex --scope user

A healthy result confirms 26 installed Skills, the manifest, current source/core bindings, and the host-specific Agent Profile expectation. Doctor reports an exact historical 27-Skill recommended, 40-Skill full, or 190-Skill dev manifest as migration-required; run upgrade without uninstalling first. Repository doctor evidence does not prove that the real Host loaded the files.

OpenAI API Zip Path

Preview and then generate the fixed Runtime bundle set:

python3 scripts/quickstart.py --agent openai-api --dry-run
python3 scripts/quickstart.py --agent openai-api

The second command builds and validates one zip per top-level Skill under the compatibility path dist/openai-api/zips/recommended/. It does not install files or run runtime doctor. See OpenAI API zip output for the direct build/validation commands and package boundary.

Submit A First Task

Slash Skill syntax is /skill-name. Open a small test repository in the configured host. Start with /engineering-control-plane, then replace the example path and command with real values:

/engineering-control-plane

Goal: Add an empty-string check to `src/example.py` without changing its public API.
Acceptance: Empty input returns the existing validation error; valid input is unchanged.
Allowed scope: `src/example.py` and its existing test file only.
Verify: Run `python3 -m unittest tests.test_example`.
Stop if ownership, contract risk, or verification differs from this request.

Some hosts do not provide native Slash UI or autocomplete. Put the literal /engineering-control-plane in the request text in that case. It expresses routing intent; it does not prove native Slash support.

Expected outcome: the main profile chooses the Direct Task path and fixes one Primary Professional route. The task agent receives that route plus zero to three selector-chosen Layer 3 items and necessary Targeted References; it does not globally reroute or preload a catalog. A task agent implements and validates the latest edit, and a separate review agent inspects the actual diff. The final handoff records changed files, commands and results, freshness, unverified scope, and residual risk. Unknown ownership or material risk switches to Analyzed Work before editing.

Continue with Usage for three request patterns and expected decision points. If setup fails, use Installation troubleshooting instead of adding --force without inspecting the conflict.