nodecode-devtune
DevTune: what AI answers say about a product, and which tools coding agents pick. A Nodecode rewrite of devtunehq/skills by devtunehq.
Get it
Tell the agent "add the devtune cell", or run nodecode add nc://devtune.
Set it up
Make a project API key under API Keys in the DevTune account sidebar, then paste it into api_key, or leave that empty and let the cell read the variable api_key_env names (DEVTUNE_API_KEY by default).
{ "devtune": { "api_key": "your-key", "window_days": 30 } }
| Member | Default | Notes |
|---|---|---|
api_key |
— | the project API key; one of api_key or a set api_key_env is required |
api_key_env |
DEVTUNE_API_KEY |
the variable read when api_key is empty |
project_id |
empty | the project to read; empty, the key's own connection answers it |
base_url |
https://devtune.ai/api/v2 |
the REST root; only a proxy or a mirror needs another |
window_days |
30 |
the measurement window a read that names none asks for: 30 or 90 |
timeout_seconds |
30 |
seconds one read may take |
A section that names no key refuses the start and says so, and the tool is not
registered until one is set. Then (restart-cells "nodecode-devtune"), or set it in
/settings. Say /devtune to see which project the key reaches and what it can read.
Use it
Ask what AI answer engines say about your product, who they cite instead, what AI crawls and refers to your site, what to fix next, or whether a fix that shipped moved the numbers. Say "which tool would Claude Code pick to add auth?" and the model reads the Agent Picks reports instead.
The model calls devtune_read with one report name; the answer is DevTune's own JSON
under one line naming what was read. From eval, the same call:
(devtune:reports) ; every report this cell reads (devtune:report :connection) ; the project the key reaches (devtune:report :visibility-summary) ; the section's window (devtune:report :visibility-diff :window-days 90) ; and how it moved (devtune:report :actions) ; what to fix next, ranked (devtune:report :action-brief :id "act_123") ; one action's brief (devtune:report :agent-picks-summary :setup "auth") (devtune:report :agent-picks-runs :cursor "c_2") ; the page after the last one
| The question | The report |
|---|---|
| How visible the product is in AI answers | visibility-summary, visibility-diff |
| Who gets cited instead | competitive-position, citation-domains, citation-pages |
| What an AI platform actually said | visibility-responses, then visibility-response |
| Topics a competitor is cited for and you are not | content-gaps, then content-gap-evidence |
| AI crawlers, answer fetches, AI referrals | traffic-summary, traffic-platforms, ai-referrals |
| Pages AI keeps asking for that do not exist | 404-demand |
| What to fix next | actions, then action-brief |
| Site readiness for AI | audit-summary, audit-findings |
| Whether shipped fixes worked | outcomes-ledger |
| Which tool coding agents pick | agent-picks-summary, agent-picks-cells, agent-picks-history |
| Individual agent runs, and what to fix | agent-picks-runs, agent-picks-run-trace, agent-picks-actions |
| A test's status, its schedule, what to set up | agent-picks-test, agent-picks-schedule, agent-picks-catalogue |
(help :devtune) answers the guidance the two skills carried, and every request
carries one line of it: brand awareness, citation presence and share of voice are
percentages from 0 to 100 while Agent Picks rates are fractions from 0 to 1, a
citation is not a mention, a null was not measured rather than zero, and an outcome
shows what moved after a fix, not that the fix caused it. *primer* is the text, so a
layer can reword it.
This cell reads. It creates, updates, archives and deletes nothing, starts no test and spends no DevTune credit, so no approval is needed before a read and a model cannot change a project through it. The write and credit-spending calls of the original's hosted MCP server are not here.
License
MIT. The original is Copyright (c) 2026 DevTune: see LICENSE and NOTICE.
