Thirteen of the fourteen skills need no Engine, Instance, manifest, or CLI. A CSV is enough for the ones that work from data. setup-aftergrid is the one skill that needs the Engine, because setting it up is its job.
By skill
| Skill | The least you can start with | Engine | Who starts it |
|---|---|---|---|
| ask-aftergrid | A description of the work. No data. | None | You start it |
| grill-question | An unfinished question. No data. | Optional | You start it |
| explore-data | A CSV, a pasted sample, or a connection you already have. | None | You or your agent |
| define-metric | A metric name, plus a CSV or a written policy. | Optional | You or your agent |
| plan-analysis | A question, plus a CSV, schema, or data map. | None | You or your agent |
| analyze | A question, plus a CSV, notebook, or connection. | Optional | You start it |
| diagnose-change | A CSV covering both periods. | Optional | You or your agent |
| write-finding | Results and notes to write up. | Optional | Your agent uses it |
| iterate-visual | A chart, an image of one, or its code. | Optional | Your agent uses it |
| make-the-call | A result and the goal it serves. | None | You or your agent |
| analysis-review | The memo, chart, notebook, or SQL to challenge. | Optional | Your agent uses it |
| revise-finding | A memo, notebook, or chart, plus the feedback. | Optional | You start it |
| learn-from-analysis | A finished analysis or correction. | None | You or your agent |
| setup-aftergrid | A repository and the complete aftergrid toolkit. | Required | You start it |
"Optional" means the skill has an Engine path you only reach by asking for a Finding or supplying a Finding directory. That path needs the complete aftergrid toolkit, which a single installed skill doesn't include. "None" means the skill doesn't use the Engine. "Who starts it" comes from each skill's frontmatter.
By agent
Checked on 2026-11-02 with skills 1.7.4, in new scratch projects, running each agent headless. These results show what was installed and loaded on that date. They don't show how well any agent does the analysis.
| Claude Code 2.1.312 | Codex CLI 0.165.0 | Cursor | |
|---|---|---|---|
Where npx skills add --agent put the files |
.claude/skills/ |
.agents/skills/ |
.agents/skills/ |
| Install prompt below | Worked | Failed in the workspace-write sandbox (EPERM on .agents/); worked with full access |
Not run; cursor-agent wasn't signed in |
Explicit-only skill (grill-question) |
Left out of the agent's skill list; /grill-question loaded it |
Left out when the agent listed skills it could use on its own; $grill-question loaded it |
Not run |
Agent-started skill (analysis-review) |
In the agent's skill list; /analysis-review returned nothing in headless mode, so ask in plain language |
The agent reported it could use it on its own | Not run |
Skills that allow both (diagnose-change) |
In the agent's skill list; /diagnose-change loaded it |
The agent reported it could use it on its own | Not run |
For Cursor, only the file copy was checked. Other agents the installer supports weren't checked.
The installer also writes skills-lock.json to the project. Each skill it copies holds SKILL.md, a small agents/openai.yaml, and any references/. Files that land in a project aren't proof that an agent read them.
Install prompt
Paste this into your agent:
Install the aftergrid analytics skills into this project.
Run this, replacing <agent> with claude-code, codex, or cursor, whichever you are:
npx skills add patrickjmorris/aftergrid --skill diagnose-change make-the-call --agent <agent> --yes
Then list every file and directory the command created, and say whether I need to restart you before the skills appear. Do not run the skills or change anything else.
If your agent's sandbox blocks the write, run the command yourself in a terminal.
Claude Code plugin
/plugin marketplace add patrickjmorris/aftergrid
/plugin install aftergrid@aftergrid
This installs all fourteen skills. Checked on 2026-11-02 with the matching claude plugin marketplace add and claude plugin install commands, from GitHub, in a clean config directory; the interactive /plugin menu wasn't run. Claude Code reported the plugin enabled with 14 skills. Loaded with --plugin-dir, /diagnose-change and /aftergrid:diagnose-change both worked in headless mode.
The plugin doesn't install the Engine's packages, so the Engine CLI still needs a checkout.