Person 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 — read this first, an external metered service is required to use the skill (not to install it): installing copies one instruction file only and needs no account, no key and nothing else first. Using the skill afterwards requires the remote Arcmira MCP server connected in my agent host at https://mcp.arcmira.com/mcp, signed in to an Arcmira account: the host signs me in through OAuth on first use, or, for a client that cannot do OAuth, an account key (arc_sk_...) from arcmira.com sent as a Bearer header instead. Arcmira reads are metered: they use credits from the account's plan first, then the on-demand budget I set in the Arcmira dashboard, and a plan or a budget can block a read (paid_plan_required, quota_exceeded, spend_limit_exceeded and similar errors). Known rates from the repository README: arcmira_describe never bills, momentum costs 10 credits a call, a search costs 4 credits per passage it returns past the first 5, and a captions transcript read costs 4 credits per started 15 minutes of its window. Do not connect the MCP server, create an account, sign in, or ask me for a key during this installation: that setup is my decision, later. Install "person-research" for me. It researches one person across the podcasts and YouTube shows Arcmira indexes, for interview, podcast-booking or meeting prep, through three lenses in one worked program run with the Arcmira MCP tool arcmira_execute_read: momentum (an attention verdict, mentions in the last 30 days against the prior 30 days, and the top shows that mention them), mentions in "appearances" mode (the episodes they were on, last 90 days, up to 40), and search — with speakerIds for the person's own words (last 180 days, my topic as the query, up to 5 passages) and with about for what others said about them (last 90 days, up to 3 passages). Names are resolved to ids first, because filters take ids only (ent_...) and a name where an id belongs throws id_required: a "suggested" pick is an assumption the answer must name with its evidence, and when several entities fit ("ask"), its options go back to me and nothing else runs. If the speaker search returns nothing, the program reads a roughly 95-second transcript window of their newest appearance instead and says those lines come from the appearance, because caption lines do not name the speaker. Repository: https://github.com/arcmira/mcp (the skill is the whole plugins/arcmira/skills/person-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/person-research/ folder from that checkout. Then compare every file in the folder with the fingerprints recorded for this listing on 2026-10-19: [FINGERPRINTS] - SKILL.md: SHA-256 eb9f275cef02933470289ba229c3fe0299a23fadc38f65779cf06fc5f9b8e435 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 file: - The skill folder holds 1 file only: SKILL.md (6,205 bytes). There are no scripts and no code in the folder, and nothing in it executes locally. As a file it is self-contained, but functionally it depends entirely on the external Arcmira MCP server (the tools arcmira_execute_read and arcmira_describe, whose current method reference arcmira_describe returns at use time, plus arcmira_feedback for the closing feedback line): that server is not in this folder and is not installed by this install. SKILL.md reads no other skill file; it only names the company-watch skill as the one that sets up monitors. - 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 during installation, no MCP server is connected, no sign-in happens, no key is asked for, and there is no telemetry in the folder. - What leaves this machine at use time, disclosed here before it ever happens, and using the skill is a separate step that only starts when I ask you to research a person: my agent sends the whole JavaScript program to the remote Arcmira MCP server at https://mcp.arcmira.com/mcp, which reads the Arcmira API (api.arcmira.com) on my account. The program carries the exact name I gave, my own words about that person as context (never your guess: a bare first name goes with no context), my topic word or phrase as the search query (the worked program's placeholder query "AI" is replaced by my topic), the resolved entity id, and the date windows (90 days for appearances and for what others said, 180 days for the person's own words). The calls in the worked program are: resolve (type "person"), momentum, mentions in appearances mode, two searches (speakerIds, then about), and, only when the speaker search is empty, one transcript window read of the newest appearance (from 5 seconds before the mention to 90 seconds after it). My host adds my credential (OAuth token, or my arc_sk_ account key as a Bearer header) when it connects: the program itself never holds the key. The person's name and my context words are therefore sent to Arcmira on every research turn — do not research anyone with this skill unless I asked for exactly that person. - The logging the repository README discloses for that server ("What we log"), which applies at use time: each tool call is logged to Arcmira's product analytics (PostHog), linked to my account, holding the tool name, my host's name and version, the server version, how long the call took, the input (the program text of an execute call, first 4,000 characters; the arcmira_describe topic; or a feedback category and note), an optional intent of at most 300 characters (my request in a few words) when my host sends one, and the outcome, not the result: ok or the error code, the result size, whether it was truncated, and the API routes the call made (for example GET /v1/search, not its query string). Before the record is stored, anything that looks like a credential or an email address is replaced with [redacted]; results, transcript text and logs printed by a program are never logged, and a call with no valid credential is not logged. - The money at use time, from the plan first and then my on-demand budget: arcmira_describe never bills; momentum costs 10 credits a call; each search costs 4 credits per passage it returns past the first 5; the fallback transcript window is a captions read at 4 credits per started 15 minutes of its window (the roughly 95-second window is one started unit). No tool takes a spending amount, and a read the plan or budget blocks comes back as an error naming the limit. The closing line of SKILL.md about monitors (each alert uses 25 credits) is not part of this skill: setting up a monitor is the company-watch skill through arcmira_execute_write, and this skill creates no monitor and runs no write tool. - The feedback line at the end of SKILL.md: if anything was wrong, slow or missing, or I want a source or capability Arcmira lacks, send one arcmira_feedback — that sends a category and a note saying what happened (optionally naming the failed call by its call_id) to Arcmira, where it is stored as a record (POST /v1/feedback). Never send one during installation, and never send one on your own initiative outside a research turn I asked for. - The honesty limits in SKILL.md: mentions and momentum count the shows Arcmira indexes, not all media; speaker tags cover only part of the index, and a chunk found with speakerIds can include other people's lines — quote only the person's own lines (lines start with "Name: " when the speaker is known), and when no line names a speaker, say the attribution covers the passage, not each line. - Do not run the worked program, do not call arcmira_describe, arcmira_execute_read or arcmira_feedback, do not connect the Arcmira MCP server and do not sign in or create an account during this install: all of that is setup and use, my decision, later, when I actually ask for a person to be researched. 3. When I say "yes, install", install it as a skill: copy the whole person-research folder into the agent's skills directory, in a folder named "person-research", keeping its 1 file (SKILL.md at its root). A manual clone followed by copying the plugins/arcmira/skills/person-research folder into ~/.claude/skills/ is the route described here; any other route is fine only as long as the installed file matches its fingerprint. 4. Verify with no network calls: frontmatter valid (name: person-research), the 1 file in place with a matching fingerprint. Do not call any Arcmira tool to verify. 5. Report what was installed, where, and what I still need to do myself (for example: connect the Arcmira MCP server myself at https://mcp.arcmira.com/mcp in my host and sign in through OAuth, or set my own arc_sk_ account key as a Bearer header; expect metered credits from my plan and then my on-demand budget when I use it, with momentum at 10 credits a call and searches at 4 credits per passage past the first 5; using the skill is a separate step that only starts when I ask for a person to be researched, and the use-time sends disclosed above — my program with the person's name and my context words to mcp.arcmira.com, and Arcmira's per-call PostHog logging of the program text — happen only then). 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.