Menu
Use only when the user explicitly asks for deep research.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
| name | deep-research |
| description | Use only when the user explicitly asks for deep research. |
Use this skill only when the user explicitly asks for deep research.
This skill is provider-agnostic. Choose the provider or implementation that best fits the user's request and the available environment.
Typical triggers:
Do not use this skill for:
web_search can answer quicklyIf the user does not specify a provider, use the bundled default implementation in scripts/parallel_deep_research.py.
The bundled script:
$ZDX_ARTIFACT_DIRIf the user explicitly wants another provider for deep research, follow that provider instead of the bundled default implementation.
Good prompts are specific about:
Good example:
Create a research report on the current landscape of developer-focused AI coding agents in 2026. Compare product positioning, core workflows, pricing signals, platform support, and notable technical differentiators. Focus on official product pages, docs, benchmark posts, and credible reporting. End with a concise competitive summary.
Weak example:
Research AI coding tools.
python3 scripts/parallel_deep_research.py \
--save-artifacts \
-- "Create a research report on the current landscape of developer-focused AI coding agents in 2026."
The bundled default implementation currently uses:
pro-fasttextAlways prefer the -fast variant when using these processors.
Do not switch away from pro-fast unless the user explicitly asks.
Keep the default implementation simple unless the user explicitly asks for a different provider or a higher-quality/slower run.
Use these as mental guidance only. The bundled default implementation stays on pro-fast unless the user explicitly asks to change it.
core-fast: lighter and cheaper; better for more structured or narrower researchpro-fast: default choice; best general option for open-ended deep researchultra-fast: stronger and more expensive; use only when the user explicitly wants deeper researchAlways prefer the fast variant for these processors.
When $ZDX_ARTIFACT_DIR is available, prefer --save-artifacts. This stores:
Artifacts should stay under $ZDX_ARTIFACT_DIR/deep-research/.
The script requires:
python3PARALLEL_API_KEY in the environmentIf PARALLEL_API_KEY is missing, stop and tell the user exactly that.
For this skill, use polling by default.
Do not build webhook infrastructure or SSE streaming unless the user explicitly asks for it. This skill is designed for occasional local/manual use, so polling is the simplest and most reliable default.
After the script completes:
If the user wants the full raw result, point them to the saved artifact or return the report directly.
Create a ship-first (ship-shaped) implementation plan with demoable MVP phases. Use when a user asks for an implementation plan, MVP plan, shipping plan, or wants to break down a feature into incremental, demoable phases. Emphasizes user journey order and daily-usable increments over polished completeness.
Cut a new zdx release. Use when the user asks to release, cut a release, ship a version, publish a build, bump the version, or generate release notes / changelog for zdx. Orchestrates version bump, changelog, and dispatching the manual GitHub Release workflow.
Add, update, or remove models and providers in the zdx LLM registry. Use when the user asks to add a new model, add a new provider, update model pricing or context limits, regenerate default models/config, or says things like "add support for X model", "add the new Y provider", "update pricing for Z", "regenerate models", or "run models update".
Explore options and pick a direction before committing to a goal or plan. Use when the user wants to think through approaches, weigh trade-offs, or decide between alternatives — not yet ready to define success criteria or slice work. Hard rule: no code, files, or implementation until the user explicitly approves a direction.
Shape a fuzzy intention into a concrete, measurable goal with explicit success evidence before starting work. Use when the user asks to define a goal, clarify success criteria, sharpen an objective, or turn "make X better" into something verifiable. Useful as a quick pre-step before planning, debugging, performance work, research, or operations work.
Use for memory-related tasks: saved notes, factual questions that may already be documented, and saving durable information. Prefer Memory_Search and Memory_Get for discovery; use this skill for routing, note-saving, and filing conventions.