Skip to content

datarobot-oss/datarobot-agent-skills / skills/datarobot-agent-assist/agent-assist-build

datarobot-agent-assist-build

Flagged

Use when the user wants to design, build, code, or deploy an AI agent on DataRobot; mentions agent_spec.yaml, dress rehearsal, the DataRobot agent template, LangGraph, CrewAI, LlamaIndex, NAT, Base agents, MCP servers, backend APIs, custom frontends, or the DataRobot CLI.

references/pre-coding-checklist.md

View on GitHub
## Pre-coding Checklist

Run this checklist when the user enters **[2. Coding an AI Agent](../SKILL.md#2-coding-an-ai-agent)** — including when offered from a failed pre-deployment check.

**On Windows: coding is not supported — stop and do not proceed** (see [SKILL.md §2](../SKILL.md#2-coding-an-ai-agent)).

**Design → code:** If the user chose **Code the agent** from [Post-design next steps](../SKILL.md#post-design-next-steps), `<design_to_code>` must be true before starting this checklist.

### Entry paths (Bootstrap skip rules)

| Entry | Skip Bootstrap | Start at |
|-------|----------------|----------|
| **Same-session design → code** (`<design_to_code>` is true) | Steps 1–2 | Template setup step 3 |
| **Code after workspace resolution** (`<workspace_resolved>` is true and `<target_dir>` equals `<workspace_resolved_target_dir>`) | Step 1 only | Step 2 (cold Code — field validation and missing-spec recovery) |
| **Deploy → coding handoff** (`<target_dir>` set, template not verified) | Step 1 only | Step 2 (spec validation and missing-spec recovery), then Template setup step 3 |

---

### Bootstrap

1. **Confirm `<target_dir>`** — the project root for this session (set in [Workspace Resolution](workspace-resolution.md)). All paths and scripts use this directory.

   **Skip** when `<workspace_resolved>` is true and `<target_dir>` equals `<workspace_resolved_target_dir>`, or when `<design_to_code>` is true, or on deploy → coding handoff.

2. **Confirm spec** — read `<target_dir>/agent_spec.yaml`, falling back to `<target_dir>/agent_spec.md` only if the YAML file is absent. If both exist, use the YAML file. Any spec changes must be written to `agent_spec.yaml`.

   **Skip entirely** when `<design_to_code>` is true (design already produced the spec).

   **Required** on deploy → coding handoff — pre-deployment does not check for `agent_spec.yaml`, but coding needs a [spec complete](resume-design.md#spec-complete) file before template setup (including `model` for step 9).

   **If the file does not exist**, do not assume Design is required — `<target_dir>` may be wrong.

   - If `<workspace_resolved>` is true and `<target_dir>` equals `<workspace_resolved_target_dir>` (workspace just set this path), ask (exact wording):

     > No `agent_spec.yaml` found at `<target_dir>` (the path you chose earlier). What would you like to do?
     > 1. **Run Design phase** — create a new spec
     > 2. **Try a different directory**

     Do **not** repeat the workspace-resolution directory question verbatim.

   - Otherwise (no recent workspace resolution for this path), ask (exact wording):

     > No `agent_spec.yaml` found in `<target_dir>`. What would you like to do?
     > 1. **Run Design phase** — create a new spec
     > 2. **Spec is elsewhere** — point to the directory that contains `agent_spec.yaml`

   Wait for the user's choice.

   - **Choice 1** (or "design" / "create") → follow **[1. Designing an AI Agent](../SKILL.md#1-designing-an-ai-agent)**. Stop the coding checklist.
   - **Choice 2** (or "elsewhere" / "subdirectory" / "different directory" / "try a different directory") → follow [Path resolution](workspace-resolution.md#path-resolution), then re-run **step 2** on the new `<target_dir>` (field validation only — do not re-run step 1).

   **If the file exists** (cold Code entry only), **validate** the spec before anything in [Template setup](#template-setup). Use [resume-design.md § Spec complete](resume-design.md#spec-complete) — this is a field check only; do **not** enter [Resume Design](resume-design.md) unless the user chooses to edit via [Spec issues](#spec-issues) below.

   **Validation procedure (required — do not skip):**

   1. After reading `<target_dir>/agent_spec.yaml`, check each required field in order: `system_prompt`, `model`, `frontend.type`, `tools` (key present). If `model` is the `datarobot-deployed-llm` placeholder, `llm_deployment_id` is required too.
   2. **Tell the user the result** in your next message — pass or fail, and list any missing fields (e.g. *"Spec validation failed: missing `model`."* or *"Spec validation passed — all required fields are present."*).
   3. **If all fields pass** → continue to [Template setup](#template-setup) step 3.
   4. **If any field fails** → follow **[Spec issues](#spec-issues)** below. **Stop** — do not classify the workspace, run `ls`, clone, or start template setup until the spec is valid or the user has resolved the issue.

   If the spec is not complete, follow **[Spec issues](#spec-issues)** below. Do not start template setup or coding until the spec is valid.

### Spec issues

When step 2 finds a problem in `agent_spec.yaml`:

1. Explain the issue briefly (what is missing or invalid).

2. **If the only gap is `tools`** (all other required fields are present), ask (exact wording):

   > The `agent_spec.yaml` is missing a `tools` definition. Would you like to:
   > 1. **Edit the spec in the Design phase**
   > 2. **Confirm no tools needed** — proceed to coding without tools

   - **Choice 1** (or edit / add tools / fix spec) → set `<design_to_code>` = false. Read and follow [resume-design.md](resume-design.md) through Agent Simulation and Post-design next steps — do not return to coding until the user chooses **Code the agent**. Stop the coding checklist.
   - **Choice 2** (or "no tools" / "none needed") → write `tools: []` to `<target_dir>/agent_spec.yaml`, then treat `tools` as satisfied. Continue to Template setup step 3.

   **Otherwise** (any other missing or invalid field), ask (exact wording):

   > `agent_spec.yaml` needs changes before coding can continue. Would you like to edit the spec in the Design phase?

   - **If yes** (or the user asks to add, fix, or update the spec) → set `<design_to_code>` = false. Read and follow [resume-design.md](resume-design.md) through Agent Simulation and Post-design next steps — do not return to coding until the user chooses **Code the agent**. Stop the coding checklist.
   - **If no** → do not proceed to template setup. Re-ask or return to the [welcome menu](../SKILL.md#on-activation).

---

### Template setup

**Prerequisite:** All [spec complete](resume-design.md#spec-complete) fields are present in `<target_dir>/agent_spec.yaml` — validated in Bootstrap step 2 (cold Code and deploy → coding handoff), or satisfied by same-session design (`<design_to_code>` is true; step 2 skipped). If step 2 reported missing fields, do **not** enter this section — handle [Spec issues](#spec-issues) first.

3. **Read `REPO_URL`** from `REPO_URL` in `<skill_scripts_dir>/clone_template.py`. This is the only canonical template repository URL — use it for remote comparison and cloning. See [helper-scripts.md](helper-scripts.md) for script details.

### Clone discipline

- **No separate clone approval** when classification is **Spec-only** (step 6) — the user already chose to code. Give a brief notice (1–2 sentences), then clone without waiting.
- **Subdirectory confirmation required** when classification is **Everything else** (step 7). Choosing **Code the agent** from [Post-design next steps](../SKILL.md#post-design-next-steps), saying "implement", or `<design_to_code>` being true does **not** satisfy this gate — wait for an explicit reply to step 7b (or 7c if the subdirectory already exists).
- **Destructive actions** (clearing an existing subdirectory) always require explicit confirmation in step 7c before proceeding.
- **Before every clone**, state what will happen: target directory, that the DataRobot agent template will be cloned there, and (for step 7) that the recognized spec file(s) will be moved. This notice does not require a reply in step 6; in step 7, the reply must come **before** the notice and clone.

4. **Classify `<target_dir>`** (evaluate in order; first match wins):

   **Design-phase artifacts (allowed in Spec-only):** dress rehearsal may leave these before coding — they do **not** count as a messy workspace:
   - `agent_spec.yaml`, legacy `agent_spec.md`, `.env`
   - `.datarobot/rehearsal/` (session cache only)
   - `rehearsal_report/rehearsal_report.md` (and the `rehearsal_report/` directory that contains it)

   Any other path under `<target_dir>` — including `.datarobot/answers/`, swarm dirs, source code, or unrelated files — still classifies as **Everything else**.

   | Classification | Conditions |
   |----------------|------------|
   | **Existing template** | Git repository, `origin` matches `REPO_URL`, `AGENTS.md` present |
   | **Spec-only** | Not existing template, and every file/directory in `<target_dir>` is one of the design-phase artifacts above (nothing else) |
   | **Everything else** | Any other state — including wrong git remote, git repo without `AGENTS.md`, or extra files/directories (e.g. `src/`, `.datarobot/answers/`, `.gitignore`) |

   **Deploy → coding handoff:** After classifying, jump to the matching step below — do not re-run Bootstrap step 1.

5. **Existing template** — notify the user the template is already present in `<target_dir>`. Then:

   a. If `.datarobot/answers/agent-agent.yml` contains `agent_template_framework` → skip framework selection (step 8).

   b. If `.env` exists (setup was run previously) → skip `setup_template.py` (step 9).

   c. Continue to step 10 (dependency check).

6. **Spec-only** — clone and set up the template in `<target_dir>`:

   a. **Move recognized spec files aside** if present — move `agent_spec.yaml` and/or the legacy `agent_spec.md` to temporary locations before cloning so they are not overwritten. Restore them to `<target_dir>` after cloning completes. Leave design-phase rehearsal artifacts in place (`.datarobot/rehearsal/`, `rehearsal_report/`) — do not move or delete them.

   b. **Notify, then clone** — tell the user (1–2 sentences, no wait): *"Your workspace is spec-only. I'll clone the DataRobot agent template into `<target_dir>` now."* Then run [clone_template.py](helper-scripts.md#clone_templatepy).

   c. Continue to framework selection (step 8).

7. **Everything else** — conflicting workspace. Do not clone into `<target_dir>` as-is.

   **STOP. Do NOT create subdirectories, move files, or run `clone_template.py` until the user has replied to step 7b or 7c.**

   a. Explain what was found in `<target_dir>`.

   b. If `<design_messy_cwd>` is true, say: *"During design you were warned this directory has other files."* Then offer the default subdirectory `new-datarobot-agent` (or another name the user provides). Tell the user that any recognized spec file(s) will be **moved** into that subdirectory (not copied). Ask the user to confirm or provide a different subdirectory name.

   Otherwise, offer to create the agent in a subdirectory (default name: `new-datarobot-agent`). Tell the user that any recognized spec file(s) will be **moved** into that subdirectory (not copied) so there is a single project location. Ask the user to confirm or provide a different subdirectory name.

   Wait for the user's reply. Do **not** treat **Code the agent**, "implement", or `<design_to_code>` as confirmation here.

   c. If the subdirectory already exists: warn that using it will **clear everything** in that subdirectory. Ask for confirmation. If the user declines, ask for a different name or return to the [welcome menu](../SKILL.md#on-activation).

   d. If the user agrees:

      - **Move** (do not copy) `<target_dir>/agent_spec.yaml` and/or `<target_dir>/agent_spec.md` into the subdirectory if present. The parent must not keep a duplicate — a stale cwd spec breaks future sessions. If both exist, keep both and use `agent_spec.yaml` as authoritative.
      - Clear the subdirectory contents if it already exists.
      - Set `<target_dir>` to the subdirectory (automatic update — do not ask the user to reset `<target_dir>`).
      - Tell the user (1–2 sentences): *"I'll clone the DataRobot agent template into `<target_dir>` now."* Then run clone (step 6b), framework selection (step 8), setup (step 9), and dependency check (step 10) in the new `<target_dir>`.

   e. If the user declines: show the [welcome menu](../SKILL.md#on-activation). Do not modify `<target_dir>`.

8. **Framework selection** (skip if step 5a applied):

   **STOP. Do NOT proceed until the user has replied with their framework choice.**

   Ask the user (exact message):

   > Which agentic framework would you like to use?
   > 1. LangGraph
   > 2. CrewAI
   > 3. LlamaIndex
   > 4. NeMo Agent Toolkit (NAT)
   > 5. Base

   Wait for the user's reply. Do not assume or default to any framework. If their next message is not a framework choice (silence, unrelated text), re-display the options and wait again — do not proceed with any other coding step. Once the user replies, map their choice to the corresponding `--framework` value (see [select_framework.py](helper-scripts.md#select_frameworkpy)) and run that script.

   Invalidate `<dependency_check_passed>` after this step.

9. **Setup** (skip if step 5b applied) — run [setup_template.py](helper-scripts.md#setup_templatepy). Use the `model` field from `agent_spec.yaml` as `--llm-model` (must be present — validated in Bootstrap step 2 for cold Code and deploy handoff, or produced by same-session design). If the spec also carries `llm_deployment_id`, pass it as `--llm-deployment-id`; it selects a DataRobot-deployed LLM, and the script refuses the deployed-LLM placeholder model without it.

   Invalidate `<dependency_check_passed>` after this step.

10. **Validate dependencies** — run after setup completes. Read and follow [dependency-validation.md](dependency-validation.md) end to end.

11. **Re-read `<target_dir>/AGENTS.md`** now that the template is ready.

12. **Recreate the TODO list** based on `agent_spec.yaml` — break down the implementation into discrete steps and add them to the TodoWrite tool.

**CRITICAL**: If any helper script fails, do **not** proceed with coding. Return the error message to the user and ask how they want to proceed.
writing-skillsobra/superpowersUse when creating new skills, editing existing skills, or verifying skills work before deploymentStars: 323kUpdated Claude CodeCodexVerifiedInstall
xlsxanthropics/skillsUse this skill any time a spreadsheet file is the primary input or output. This means any task where the user wants to: open, read, edit, or fix an existing .xlsx, .xlsm, .xltx, .csv, or .tsv file (e.g., adding columns, computing formulas, formatting, charting, cleaning messy data); create a new spreadsheet from scratch or from other data sources; or convert between tabular file formats. Trigger especially when the user references a spreadsheet file by name or path — even casually (like "the xlsx in my downloads") — and wants something done to it or produced from it. Also trigger for cleaning or restructuring messy tabular data files (malformed rows, misplaced headers, junk data) into proper spreadsheets. The deliverable must be a spreadsheet file. Do NOT trigger when the primary deliverable is a Word document, HTML report, standalone Python script, database pipeline, or Google Sheets API integration, even if tabular data is involved.Stars: 198kUpdated Claude CodeCodexVerifiedInstall
shipping-and-launchaddyosmani/agent-skillsPrepares production launches. Use when preparing to deploy to production, or when asking what needs to be in place before shipping. Use when you need a pre-launch checklist, when setting up monitoring, when planning a staged rollout, or when you need a rollback strategy.Stars: 119kUpdated Claude CodeCodexVerifiedInstall
childhoodthedotmack/claude-memBabysit a shipped release after merge, publication, or deployment. Verify published packages, production behavior, installed plugins, and late CI or user reports until the observation window is clean; use babysit for PR readiness before release.Stars: 118kUpdated Claude CodeCodexVerifiedInstall