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/resume-design.md

View on GitHub
## Resume Design

Resume **[1. Designing an AI Agent](../SKILL.md#1-designing-an-ai-agent)** on an existing `agent_spec.yaml` or legacy `agent_spec.md` in `<target_dir>`. Prefer YAML when both exist. If only the legacy file exists, read its complete YAML content and copy it unchanged to `agent_spec.yaml` before classifying or editing; leave the legacy file unchanged. Then classify the selected spec, start at the matching Design section, and **run that section and every section after it in order** — do not skip ahead to coding or template setup.

---

### When to use

- The user enters **Design** (menu option 1) and continues an existing spec ([workspace-resolution](workspace-resolution.md))
- The user chooses to edit the spec from [Spec issues](pre-coding-checklist.md#spec-issues) during pre-coding

**Not** Resume Design — use the [Post-design review loop](#exceptions) instead (Post-design menu option 2).

---

### Entry paths

| Entry | Before resuming |
|-------|-----------------|
| **Design menu** — continue existing spec | Set `<design_to_code>` = false |
| **Pre-coding** — Spec issues → edit spec | Set `<design_to_code>` = false; stop the [pre-coding checklist](pre-coding-checklist.md); do not edit `agent_spec.yaml` inline during pre-coding |

After Design finishes (via Agent Simulation → Post-design next steps), coding may resume only when the user chooses **Code the agent** (set `<design_to_code>` = true and re-enter the [pre-coding checklist](pre-coding-checklist.md)).

---

### Spec complete

Canonical definition used by [Bootstrap step 2](pre-coding-checklist.md#bootstrap) (cold Code entry — **validation only**, does not start Resume Design) and the resume table below. A spec is **complete** when all of the following hold:

| Field | Requirement |
|-------|-------------|
| `model` | Set (non-empty) |
| `llm_deployment_id` | Set **only if** `model` is the `datarobot-deployed-llm` placeholder; otherwise absent or empty |
| `system_prompt` | Non-empty |
| `frontend.type` | Set |
| `tools` | Key **present** in YAML — either one or more tool entries, or `tools: []` |

**`llm_deployment_id` rule:** the placeholder model is shared by every DataRobot-deployed LLM, so on its own it does not name one. A spec carrying it without an id is **not** complete — route to [Model Selection](../SKILL.md#model-selection) to re-pick the deployment. Template setup refuses that pair anyway; catching it here keeps the failure in the design phase.

**`tools` rules:**

- **Missing `tools` key** ≠ complete. Route to [Spec Display](../SKILL.md#spec-display) to add tools (or confirm none in Design).
- **`tools: []`** is valid only after the user **explicitly confirms no tools are needed** — during Design or via pre-coding [Spec issues](pre-coding-checklist.md#spec-issues) choice 2. Write `tools: []` to `agent_spec.yaml` when that confirmation happens during pre-coding.

---

### Classify spec state

Read `<target_dir>/agent_spec.yaml` after applying the legacy-file promotion rule above, and evaluate **top to bottom**; use the **first matching row** in the [resume table](#resume-table). If multiple gaps exist, the first row still wins (earliest Design section).

| Check | Treat as |
|-------|----------|
| File empty or no meaningful YAML content | Spec empty |
| `system_prompt` key missing, or value empty / whitespace-only | Empty or missing `system_prompt` |
| `model` key missing or empty | Missing `model` |
| `model` is the `datarobot-deployed-llm` placeholder and `llm_deployment_id` is missing or empty | Missing `llm_deployment_id` |
| `frontend.type` not set | Missing `frontend.type` |
| `tools` key absent | Missing `tools` |
| All [spec complete](#spec-complete) requirements met | Complete |

---

### Resume table

| State of `agent_spec.yaml` | Start at | Sections that follow (in order) |
|--------------------------|----------|----------------------------------|
| Spec empty | [Clarification Phase](../SKILL.md#clarification-phase) | Model Selection → Frontend Check → Spec Display → Agent Simulation → Post-design next steps |
| Empty or missing `system_prompt` | [Clarification Phase](../SKILL.md#clarification-phase) | Model Selection → Frontend Check → Spec Display → Agent Simulation → Post-design next steps |
| Missing `model` only | [Model Selection](../SKILL.md#model-selection) | Frontend Check → Spec Display → Agent Simulation → Post-design next steps |
| Missing `llm_deployment_id` (placeholder `model`, no id) | [Model Selection](../SKILL.md#model-selection) | Frontend Check → Spec Display → Agent Simulation → Post-design next steps |
| Missing `frontend.type` only | [Frontend Check](../SKILL.md#frontend-check) | Spec Display → Agent Simulation → Post-design next steps |
| Missing `tools` only (`tools` key absent) | [Spec Display](../SKILL.md#spec-display) | Agent Simulation → Post-design next steps |
| [Complete](#spec-complete) | [Spec Display](../SKILL.md#spec-display) | Agent Simulation → Post-design next steps |

**Section chain rules:**

- **Frontend Check** — skip when `frontend.type` is already set in `agent_spec.yaml` (same as [SKILL.md](../SKILL.md#frontend-check)).
- **Agent Simulation** — always run after Spec Display on this path (dress rehearsal offer), unless the [Post-design review loop](#exceptions) applies.

### Spec Display → Agent Simulation (required transition)

On every Resume Design path, [Spec Display](../SKILL.md#spec-display) is **not** the end of design — even when the spec becomes [complete](#spec-complete) during that section (e.g. adding the missing `tools` key after pre-coding [Spec issues](pre-coding-checklist.md#spec-issues)).

**After showing the spec:**

- Invite refinement only (system prompt, tools, model, examples). Follow [SKILL.md § Spec Display](../SKILL.md#spec-display) — one design gate per turn.
- Do **not** ask about coding, template setup, or "proceeding to coding" in the same turn or as an alternative to refinement. Completing the spec does **not** authorize skipping to coding.

**When the user is done refining** (e.g. "looks good", "specs looks good", "no changes", "move on"):

- In your **next** response, go directly to [Agent Simulation (Before Coding)](../SKILL.md#agent-simulation-before-coding) — present the dress-rehearsal prompt using the exact wording in SKILL.md.
- Do **not** offer [Post-design next steps](../SKILL.md#post-design-next-steps) yet. The coding option appears only **after** the user declines dress rehearsal or finishes a rehearsal session.

**Wrong (do not do this on Resume Design):**

> The spec is now complete. Would you like to refine anything, or shall I proceed to coding?

**Right:** refinement invite only → user confirms → dress-rehearsal offer → (if declined) Post-design next steps menu.

---

### Exceptions

**Post-design review loop** — when the user chooses **Review / edit spec** from [Post-design next steps](../SKILL.md#post-design-next-steps): stay in [Spec Display](../SKILL.md#spec-display) only, then return to the Post-design menu. Do **not** re-run Resume Design, Clarification, Model Selection, Frontend Check, or Agent Simulation unless the user leaves and re-enters via an entry path above.
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