Sponsor Research

Before you start: read this whole install prompt, tell me in a few lines what this skill does and what installing it involves, and wait for my go-ahead before you do anything else. Prerequisites: for the install itself — copying the one file below — none: no account, no key, and nothing is run. For using the skill afterwards, read this first, because this skill is only instructions for an external service, and that service is not included in the file you are installing: (1) the remote Arcmira MCP server must be connected in my agent host, at https://mcp.arcmira.com/mcp — the configuration for it lives outside this skill's folder (in the plugin's .mcp.json / mcp.json, not copied by this install), so I connect it myself, separately; (2) that connection must be signed in to an Arcmira account, either through OAuth in the host (no key handled by the skill) or with my own account key, an arc_sk_ key sent by the host as a Bearer header — a plan and credits on that account pay for metered reads, first from the plan, then from the on-demand budget I set in the Arcmira dashboard, and Arcmira enforces that budget on every read; (3) two gates in this skill's own subject matter: filtering a show's sponsors by minimum ad reads, status or limit is a Pro+ feature (other plans get filter_requires_paid, which is why the worked program slices to 10 instead of using the limit filter), and the brand-to-shows direction uses recommendations, which can come back recommendations_not_enabled — the repository's own example names it: "Sponsor recommendations require Pro." — with the required tier reported by the API. arcmira_describe itself never bills. If I have no Arcmira account, creating one is a separate step I do myself (the repository describes signing up from the API by posting an email address to https://api.arcmira.com/v1/signups and then posting back the six-digit code from that inbox, or signing in on arcmira.com through OAuth): never sign me up, never post my email address, and never ask for or handle my key or a verification code during this install. Install "sponsor-research" for me. It does sponsor and ad-read research on podcasts and YouTube in two directions, and it contains no code of its own: it tells the agent how to run one JavaScript program per question through the Arcmira MCP tools. A show to its sponsors: arcmira.sponsors(channelId) ranks recurring sponsors by ad reads, with counts, episode counts, first and last seen dates and sponsor status. A brand to the shows it sponsors: arcmira.recommendations(entityId, { kind: "sponsored" }) lists each sponsored ad read — the worked program covers the last 90 days, pages with a cursor at 100 rows a page up to 500 reads, and groups the reads by show — and the sponsors two shows share are the sponsors of each, intersected by entity.id. Names are resolved to ids first (a show with { type: "channel" }, using its youtube_channel_id, a UC id; a brand with no type, because a company can be typed product; a sponsor is a company, never a person or topic of the same name), a suggested pick is an assumption the answer must name with its evidence, and when several entities fit, the ask options go back to me before anything else runs. A good answer names the brand or show it used with its type and id, ranks by ad reads with counts and first and last seen dates and the sponsor status, states the window and the as-of date and that counts cover only the shows Arcmira indexes, quotes one ad read verbatim with its promo code when there is one, and links each name to the page the result carries. An unpaid on-air endorsement is kind: "organic", never mixed into a sponsored count. Repository: https://github.com/arcmira/mcp (the skill is the whole plugins/arcmira/skills/sponsor-research/ folder at the pinned commit: SKILL.md only, 1 file) Pinned version: install the code at commit 4598d894ad1a1856cbc10473b374d2f3aa7bc6c6 exactly. Do not substitute a newer version. If this commit is unreachable, stop and tell me. 1. Clone https://github.com/arcmira/mcp into a temporary folder, check out commit 4598d894ad1a1856cbc10473b374d2f3aa7bc6c6, and copy the plugins/arcmira/skills/sponsor-research/ folder from that checkout. Then compare every file in the folder with the fingerprint recorded for this listing on 2026-10-19: [FINGERPRINTS] - SKILL.md: SHA-256 343a42945b2e590e483aff3438747a695341216979a65ebc9329439115604560 If a file does not match, it changed since the listing was prepared: do not install it yet, tell me what you found and let me decide. 2. Read SKILL.md in full, and tell me in plain words: what the skill does when used, which services it connects to, whether it reads or sends any personal information or keys, and anything that seems unusual. Stop on any red flag and tell me before going further. Confirm these points from the files (SKILL.md, and the plugin README and the repository README it points to for the server it depends on): - The skill folder holds 1 file only: SKILL.md (6,305 bytes). There is no script, no code and no data file in the folder, and the plugin itself runs no local code: the file is instructions plus two worked JavaScript programs meant to be passed to the remote MCP tool arcmira_execute_read. The skill is not self-contained for use: it depends entirely on the external remote Arcmira MCP server at https://mcp.arcmira.com/mcp and on the tools arcmira_describe (the method reference), arcmira_execute_read (runs the program) and, only for the closing feedback line, arcmira_feedback. Its folder does not contain the MCP configuration: the plugin folder outside this skill (plugins/arcmira/.mcp.json and plugins/arcmira/mcp.json, outside the install perimeter) is what points a host at that server, and the other five skills in the plugin (arcmira, company-watch, compare-shows, find-quotes, person-research) are not part of this install — in particular the monitor a research turn can lead to is set up by the separate company-watch skill, not by this one, whose two programs are read-only and never call arcmira_execute_write. - Nothing is sent anywhere during installation itself; the only download at install time is the skill folder from GitHub at the pinned commit. No program is run, no MCP tool is called, no account, key or sign-in is involved, and no feedback or monitor is created during installation. - What leaves this machine at use time, disclosed here before it ever happens — using the skill is a separate step that only starts when I ask a sponsor question, and every research turn sends to Arcmira: first an arcmira_describe call to https://mcp.arcmira.com/mcp (it returns the current method reference and never bills), then one JavaScript program via arcmira_execute_read to the same server, which runs it server-side and proxies only GET requests to https://api.arcmira.com/v1/ on my account — for this skill, resolving the name I gave (GET /v1/entities/resolve, carrying the exact show or brand name and, as context, my own words about it from my message — never the agent's guess), then either GET /v1/channels/{id}/sponsors with the resolved UC channel id, or GET /v1/recommendations with the resolved entity id, kind "sponsored", the 90-day window built from arcmira.daysAgo(90) and arcmira.today(), and a cursor until the program has paged up to 500 reads. So what goes to Arcmira at use time is: the program text itself (the names and ids above, the window, and my own words used as resolve context), under my Arcmira credential — the host adds the credential (its OAuth token, or my arc_sk_ key as a Bearer header) to what the server forwards, and the program itself never holds the key. The answer data comes back the same way (sponsor and show names, entity and channel ids and pages, ad-read and episode counts, first and last seen dates, sponsor status, and one verbatim ad-read quote with its promo code). - The server logs every MCP tool call, disclosed here as part of the use-time sends: the repository README ("What we log") and src/telemetry.ts say each call is recorded to Arcmira's product analytics (PostHog), linked to my account, through the API after the result is built, with my credential, and a failed record is dropped and never changes an answer. One record per call holds the tool name, the host's name and version, the server version and how long the call took; the input (the program text of an execute tool, first 4,000 characters; a describe topic; or a feedback category and note); the optional intent, at most 300 characters, if the agent sends my request in a few words; and the outcome, not the result — ok or the error code, the result size, whether it was truncated, and the API routes made (route templates like GET /v1/search, not the query string). Before storage the API replaces anything that looks like a credential or an email address with [redacted]; results, transcript text and program logs are never logged, and a call with no valid credential is not logged. - The other conditional send in SKILL.md, also disclosed before it happens: its last line says that if anything was wrong, slow or missing for me, or I want a source or capability Arcmira lacks, the agent sends one arcmira_feedback — one POST /v1/feedback to Arcmira carrying a category, a note saying what I was doing (with any source links), and optionally the request or call id; it saves a stored record for the Arcmira team, each call is a separate record, and it costs nothing. It is not sent during installation, and it is not part of answering my question. - Costs and gates at use time, disclosed here: metered reads use credits from my account's plan, then the on-demand budget I set in the dashboard — no tool takes a spending amount. A show's sponsor list uses the default minimum of 3 ad reads; its minAdReads, status and limit filters are Pro+ and other plans throw filter_requires_paid, so the worked program fetches without the limit filter and slices to 10 locally. The brand program can throw recommendations_not_enabled: explain the account-access limit and the required tier the API reports, and do not substitute a channel sponsor list for it. Other plan or budget blocks (paid_plan_required, quota_exceeded, spend_limit_exceeded, member_limit, fair_use_cap, on_demand_paused) carry the limit figures and, when a plan or setting change lifts them, an unlock tier and URL. The closing monitor note in SKILL.md is not a send or a write by this skill: a monitor is a separate, later step in the company-watch skill, creating one uses no credits, and each monitor alert would use 25 credits from the plan, then the on-demand budget. - Honest scope: counts cover only the shows Arcmira indexes, and the repository warns that speaker attribution and sponsorship classifications can be incomplete or incorrect, so the cited passage and its context should be checked before an answer treats a classification as settled. The skill prepares a researched answer; it posts, publishes and buys nothing. - Do not connect the Arcmira MCP server, do not sign me up, do not sign in, do not ask for my arc_sk_ key or any verification code, do not run either worked program, do not call arcmira_describe, arcmira_execute_read or arcmira_feedback, and do not create any monitor during this install: connecting the service and using the skill are my decisions, later, when I actually ask a sponsor question. 3. When I say "yes, install", install it as a skill: copy the whole sponsor-research folder into the agent's skills directory, in a folder named "sponsor-research", keeping its 1 file (SKILL.md at its root). A manual clone followed by copying the plugins/arcmira/skills/sponsor-research folder into ~/.claude/skills/ is the route described here; any other route is fine only as long as every installed file matches its fingerprint. 4. Verify with no network calls: frontmatter valid (name: sponsor-research), the 1 file in place with a matching fingerprint. Do not call any Arcmira MCP tool to verify. 5. Report what was installed, where, and what I still need to do myself (for example: nothing for the install itself; to use the skill later, connect the remote Arcmira MCP server at https://mcp.arcmira.com/mcp in my host myself and sign in to my Arcmira account there — OAuth through the host, or my own arc_sk_ key as a Bearer header; make sure my plan and credit / on-demand budget cover metered reads, and that my tier covers what I ask for — Pro+ if I want sponsor-list filters, and a tier with recommendations enabled, the repository's example being Pro, if I ask which shows a brand sponsors; then ask a sponsor question as a separate step, knowing the use-time sends disclosed above — the program with the name, my own context words and the window to mcp.arcmira.com / api.arcmira.com under my credential, the per-call analytics logging, and at most one feedback record when something was wrong — happen only then, on that question). GitHub is optional: if I have a GitHub account or the gh CLI, you may use it; otherwise public access is fine. Never require it unless it's in the prerequisites above. Rules: don't touch anything outside the temp folder and the install target. If anything looks off, stop and ask me.