t1k:marketing:insighttracker:creative-signal
| Field | Value |
|---|---|
| Module | insighttracker |
| Version | 0.3.7 |
| Effort | medium |
| Tools | — |
Keywords: ad impressions by game, crawl insightrackr, creative signal, daily creative signal, insight athena, insightrackr, promoted games chart, top creative per game, top playables, top promoted, top promoted games
How to invoke
Section titled “How to invoke”/t1k:marketing:insighttracker:creative-signal[--date YYYY-MM-DD] [--post] [--dry-run]Insightrackr Creative Signal
Section titled “Insightrackr Creative Signal”Read what competitors are promoting right now from Insightrackr
(insight.athena.tools): the Top Promoted game charts per window, the
best-performing creative per game, and the playables, then optionally post the
day’s report to Discord. The site has no public API and no export endpoint, so
every read goes through a signed-in Chrome profile using the browser
extension’s in-page javascript_tool, which is what makes the requests
same-origin and cookie-authenticated.
Impressions are estimates Insightrackr models, never measured figures. Say so whenever you quote a number.
When to Use
Section titled “When to Use”- “What are competitors promoting right now?” / a Top Promoted chart read
- Pulling the top creative or playable for one or more games
- Running, rerunning or debugging the daily Creative Signal job
- Adding a window, a chart type, or a new Discord section to that job
Purpose
Section titled “Purpose”This skill covers the read path only: which endpoint answers which question,
how the snippet runs inside a signed-in tab, and what goes wrong. It does NOT
cover campaign creation, ad buying, or creative upload — those belong to the
ua module; interpretation of the numbers belongs to analytics.
Executed by
Section titled “Executed by”The fetch step is a headless browser run, not an inference job. Delegate it and keep the judgement inline:
| Step | Agent |
|---|---|
| Launch the signed-in profile, open the tab, run the snippet, report the result line | t1k-fact-fetcher — or a plain claude -p --chrome run (see the daily job (references/daily-job.md)) |
| Shape the returned snapshot into a report, decide what the numbers mean | inline (reasoning tier) |
| Post the report to Discord | the job’s poster, not an agent |
t1k-fact-fetcher holds no Chrome MCP tool, so it can drive the job only
through the shell script’s claude -p --chrome invocation; a raw browser read
in the main session needs the mcp__claude-in-chrome__* tools directly.
Authentication — reuse a signed-in profile
Section titled “Authentication — reuse a signed-in profile”The site sits behind a Google login. To read it, reuse the Chrome profile that
is already signed in and launch it through t1k-chrome-profile; never create
or switch to a throwaway profile for a crawl, which forces a fresh login and
gives the run no session at all. Verify the account after launching (the
profile’s Google account, not the “Browser N” name), then work in that browser.
That skill already owns listing, resolving, launching and selecting a profile in Claude in Chrome, so this module ships no profile code of its own. It is a declared dependency:
"dependencies": { "core": ">=1.0.0", "theonekit-core:t1k-delegation": ">=2.0.13 <3.0.0"}t1k-chrome-profile lives in theonekit-core’s t1k-delegation module, which is
required: true there — so it is present in any working install, and the range
only pins the version this module was written against.
Endpoints
Section titled “Endpoints”All three are POST, same-origin, Content-Type: application/json, and answer
{ code: 200, data: { list: [...] } }. code !== 200 is an application error,
not an auth failure; a non-JSON content type is the auth failure (ERR:AUTH).
| Question | Endpoint | Key body fields |
|---|---|---|
| Top Promoted games | /cas/api/unified/app/search | baseOption.tradeLevel1: ["602"] (Games), dayMode D3/D7/D30, sortField: "15" (impressions), sortRule: "desc", gptSearch: false |
| Top creatives (video/image) | /cas/api/v2/imagevideo/search | productIds: [<ids>], keyWordType 0,1,2,3,4,6,8,9, baseOption.dayMode: "DD", startTime/endTime, sortField: "15" |
| Top playables | /cas/api/v2/preplay/search | same shape, keyWordType 0,1,2,3,4 |
Each chart row carries unifiedProductId, unifiedProductName,
unifiedCompanyName, impression, and productIds — productIds is what the
two creative endpoints need. The human page for one game is
https://insight.athena.tools/app/detail/<id>/creative?type=2.
Full request bodies, the field meanings, and the sort/size knobs: references/endpoints.md.
Running the fetch
Section titled “Running the fetch”The snippet at scripts/browser-fetch.js runs inside a signed-in
insight.athena.tools tab, never in Node: it uses fetch with
credentials: 'same-origin' (the session cookie is the auth), and it saves its
results as browser downloads with fixed names.
- Launch the signed-in profile —
t1k-chrome-profile(~/.claude/skills/t1k-chrome-profile/scripts/chrome-profile.cjs launch <selector> --url https://insight.athena.tools/). That skill is the only profile tool this module uses; it is declared as a cross-kit dependency inmodule.json, so installing this module installs it too. Never a throwaway profile: a profile with no Insightrackr session returnsERR:AUTH. - Confirm the extension is connected (
list_connected_browsersshows a device) and open a new tab at the site. - Verify the tab host after navigation. A Google sign-in host instead of
insight.athena.toolsmeans signed out — stop withERR:AUTH, do not retry. - Pass the file’s content verbatim, prefixed with
await, tomcp__claude-in-chrome__javascript_toolin that tab. - Read back only the single result line, then close the tab.
The result line is OK date=<date> games=<n> videos=<n> or ERR:<reason>.
videos= is how many video files it actually saved (a game can be image-only,
or its video can exceed the 10 MiB cap), so a caller waiting for files waits for
exactly that many.
Every step, the exact prompt text, and the file-wait logic: references/browser-fetch.md.
The daily job
Section titled “The daily job”scripts/run-daily.sh is the whole job: launch profile, wait for a connected
browser, run the snippet with retries, wait for the downloads, post to Discord.
Install the timer with scripts/install.sh, remove it with --uninstall.
Exit codes — the poster’s alert message is chosen from them:
| Code | Meaning |
|---|---|
| 0 | posted |
| 2 | setup/usage: a prerequisite file or binary is missing |
| 3 | signed in again needed — the Insightrackr session expired |
| 4 | Chrome not reachable by Claude in Chrome |
| 5 | fetch failed after retries |
| 6 | the snapshot/video files never arrived in the downloads folder |
| 7 | the Discord post failed |
INSIGHT_DRY_RUN=1 prints the plan without touching a browser or posting.
Details, log location and rerun commands: references/daily-job.md.
Vault entries
Section titled “Vault entries”| Name | Purpose |
|---|---|
TOS_INSIGHT_DISCORD_WEBHOOK | Discord webhook URL the daily job posts to |
INSIGHT_DISCORD_WEBHOOK | Optional env override of the same value |
Read with t1k vault get <name>. Never write a webhook URL, cookie, or token
into a skill file, a snapshot, a log, or a Discord message — the site session
lives in the Chrome profile, and nothing in this module needs to know it.
Hard rules
Section titled “Hard rules”- Claude in Chrome must be connected in the right profile. The extension is
per profile; the signed-in work account is the one that is normally used.
list_connected_browsersnames devices “Browser 1”, so verify the account rather than trusting the name. - Signed media URLs expire in about 5 hours. Never return one to the caller or store it: download the bytes inside the tab during the same run.
- A playable’s HTML is a tiny AppLovin loader, not the playable itself. It cannot be posted as a file; link the game page instead and mention the playable’s impression count.
- Discord house rules for anything posted: the first line is
-# ✅ No action needed, no em-dashes anywhere (data included), every message under 2000 characters, and the owner is tagged on the first message only. - Automatic downloads are throttled by Chrome back to back, so the snippet waits between saves; a caller that polls for files must allow for that.
- Videos over 10 MiB are skipped on purpose — Discord upload limits, not a
fetch failure. The snapshot records
skipped: 'over-10MiB'.
References
Section titled “References”- references/endpoints.md — request bodies and field meanings
- references/browser-fetch.md — the in-tab procedure, step by step
- references/daily-job.md — the systemd job, exit codes, logs, reruns
- references/discord-post.md — message shape and the house rules