Cookiy AI user-research-skill: an agent skill that routes research work through Cookiy AI
Cookiy AI Skill for AI agents (Claude, Codex, Cursor, OpenClaw) — end-to-end user research: AI interviews, synthetic users, quant surveys, participant recruitment.
At a glance
- What is it?
- The repository packages research planning, transcript synthesis, AI-moderated interviews and surveys as an installable skill for Claude Code, Codex, Cursor and OpenClaw. The planning and synthesis parts run locally in the agent; anything involving real or synthetic participants goes through Cookiy AI's API.
- Who is it for?
- Adopt it if your team already runs studies through Cookiy AI and wants the planning and synthesis steps to happen inside Claude Code, Codex or Cursor instead of a browser tab. Skip it if your research data cannot leave your infrastructure, or if you only need interview guides and transcript coding, since a general-purpose prompt does that without a plugin install and a network allowlist.
- Can I use it commercially?
- Yes. MIT is a permissive licence: you can use, modify and sell software built on it, as long as you keep its copyright and licence notices.
- Is it still maintained?
- Yes. The repository last received commits 27 days ago.
- What is it written in?
- Mainly Shell, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on September 15, 2026, and from our analysis. They are not legal advice.
DEEP OPEN-SOURCE ANALYSIS
The gap this skill fills between an agent and actual respondents
An AI agent can draft an interview guide in seconds. It cannot tell you what a churned customer actually thought, because it has no channel to a respondent. This repository is an attempt to close that gap: it wraps research planning, transcript synthesis, AI-moderated interviews and multi-language surveys into a single skill that Claude Code, Codex, Cursor or OpenClaw can load. The README describes it as the skill layer for agents, with the Cookiy.ai website as the human-facing product. The two are connected by account, so a study created on the website appears in the CLI skill and results stay in sync.
The audience is narrow and specific. It is for product and design researchers who already use an agent as their working environment and who are willing to route participant-facing work through a third-party service. It is not a general research library. The planning and synthesis capabilities stand alone, but the qualitative and quantitative capabilities are explicitly marked "via Cookiy AI" in the capabilities table, which means they depend on an account and a reachable API rather than on the skill files themselves.
How the skill splits local work from Cookiy AI calls
The capabilities table divides the skill into four rows, and the division matters more than the feature list. Research Planning and Report Synthesis are described as generation and transformation tasks: a plan, a screening questionnaire, an interview guide, or a structured report built from raw transcripts with codebooks, personas and prioritized findings. Those read as prompt-and-file operations that an agent can perform with the skill loaded. Qualitative Studies and Quantitative Surveys are described as running "via Cookiy AI", which implies an HTTP call to a backend that hosts the interview moderation, the participant pool and the survey engine.
The README gives one detail about the synthesis path: it is a "five-phase synthesis pipeline". That is the whole description. Five phases are named nowhere in the supplied material, so the intermediate artifacts, the ordering and whether transcripts leave the machine are all unconfirmed. The network configuration hints at an answer without settling it. Claude Cowork and Claude Chat Desktop users must allow network egress and either set the domain allowlist to all domains or add s-api.cookiy.ai as an additional allowed domain. A single API hostname in the allowlist is consistent with participant-facing calls only, but it does not prove that synthesis is local. Anyone whose transcripts contain sensitive material should read the skill files before running the pipeline, not after.
Install paths: three commands, a plugin marketplace, or npx
Claude Code Terminal users get the shortest path. The README gives three commands in sequence: /plugin marketplace add cookiy-ai/user-research-skill, then /plugin install user-research@cookiy-ai, then /reload-plugins. Auto-update is opt-in through the /plugin interface, choosing the Marketplaces tab, then cookiy-ai, then Enable auto-update. For Codex, Cursor, OpenClaw and other agents the README offers npx cookiy-ai as a single command, with a manual installation fallback that follows each agent's own skill instructions.
Claude Chat Desktop takes a different route. You download the skill ZIP from the latest release, then use Customize, Skills, the plus button, Create Skill, Upload a skill. Claude Cowork and Code Desktop use a marketplace flow instead: Customize, Personal Plugins, plus, Create Plugin, Add marketplace, enter cookiy-ai/user-research-skill, click Sync, then under Plugins and Personal click the plus next to User research. Both Claude desktop paths require the same network step: Profile, Settings, Capabilities, Code execution and file creation, turn on Allow network egress, then either set Domain allowlist to All domains or add s-api.cookiy.ai under Additional allowed domains. That last step is the one people will miss, and it fails silently in the sense that the agent will still answer planning questions while participant-facing calls break.
Invocation is either explicit or automatic. Typing /user-research-cookiy forces the skill; otherwise the README says the agent loads it by semantic matching when you describe a research goal.
Where the skill stops being the right tool
Three limits are visible in the material. The first is the network dependency. Everything participant-facing requires egress to s-api.cookiy.ai and a Cookiy AI account, so a team that cannot send research traffic to an external service gets planning and synthesis only, and even the synthesis boundary is unclear. The second is that the repository is Shell, which tells you the skill is glue: prompts, file handling and API calls. There is no local interview engine, no survey renderer and no participant pool in this repository. If Cookiy AI's service changes or the API moves, the skill's core capabilities move with it.
The third limit is methodological, and the README raises it without resolving it. Synthetic users sit in the description next to real participants. Synthetic respondents are useful for piloting an interview guide or stress-testing a survey's conditional logic before spending recruitment budget, but they cannot tell you what your users did last week. The README does not say how synthetic and real participants are distinguished in a study, in the results, or in the report. Until that is clear, treat synthetic output as a rehearsal, not as evidence, and check the study configuration before citing any finding that came from a synthetic participant.
How this differs from Lyssna, Maze and User Interviews
The obvious alternatives are research platforms such as Lyssna, Maze or User Interviews, which also handle recruitment, survey logic and interview scheduling. The difference is where the work is initiated. Those platforms assume a human in a browser: you build a study in their UI, share a link, and read results in a dashboard. Cookiy AI inverts that. The study is created from the agent conversation, and the agent is also the tool that drafts the plan and writes the report. If your workflow already lives in Claude Code or Cursor, that removes a context switch and keeps the research artifacts in the same repository as the product work.
There is a second difference in scope. A dedicated survey tool generally does not read your transcripts and produce a codebook, and a general-purpose agent does not recruit respondents. This skill sits between the two, and the cost of that position is the account requirement. A team that only needs interview guides and transcript coding gets most of the value from a plain prompt with no plugin, no marketplace entry and no domain allowlist change. The skill earns its install when the same conversation needs to reach actual respondents.
Maintenance, releases and the MIT licence
The repository is MIT licensed, which permits commercial use, modification and redistribution provided the copyright notice and licence text are retained. That is the whole of the licence implication here; anything beyond it depends on Cookiy AI's separate terms of service for the hosted product, which the README does not reproduce and which this review cannot assess.
Maintenance looks light. The latest release is dated 2026-04-16 and the last push to the default branch is 2026-08-19, so the repository has been touched since the release but no newer tagged release appears in the supplied material. The only upgrade mechanism documented is the Claude Code auto-update toggle under the Marketplaces tab, and the ZIP download for Claude Chat Desktop, which a user has to repeat by hand. If you install through npx cookiy-ai, the README does not describe an update path at all. For a skill that is mostly prompts and API calls, that is a manageable cost: the surface area is small and a breaking change would most likely appear as a failed call to s-api.cookiy.ai rather than as a subtle behavioural shift. The larger maintenance risk is version drift between the skill and the Cookiy AI backend it calls, and nothing in the repository documents a compatibility contract between them.
Editorial conclusion
Adopt it if your team already runs studies through Cookiy AI and wants the planning and synthesis steps to happen inside Claude Code, Codex or Cursor instead of a browser tab. Skip it if your research data cannot leave your infrastructure, or if you only need interview guides and transcript coding, since a general-purpose prompt does that without a plugin install and a network allowlist. Before installing, verify that s-api.cookiy.ai is reachable from your agent runtime, and read the skill's own files to confirm whether the synthesis pipeline sends transcripts to Cookiy AI or processes them locally.
Community notes