Arcmira

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 MCP service and an Arcmira account are required to USE this skill, and use is metered. Installing the skill itself is just copying one instruction file: no account, no key and no payment for the copy. But the skill contains no code and no server: it only works when my agent host is also connected to Arcmira's remote MCP server at https://mcp.arcmira.com/mcp, which is a separate setup step that stays my decision and is NOT part of this install. Connecting that server means signing in to an Arcmira account through the host (OAuth sign-in and consent on arcmira.com, a token carrying the permissions I allow, refreshed by the host), or, for a client that cannot do that sign-in, sending an Arcmira account key (arc_sk_...) as a Bearer header instead. Metered reads use credits from my account's plan first, then the on-demand budget I set in the Arcmira dashboard (no tool takes a spending amount, and Arcmira enforces that budget): a search uses 4 credits per passage it returns past the first 5, and a Premium transcript read costs far more (the repository README states 300 credits per started 15 minutes of the whole video; arcmira.quote returns that price beforehand and uses no credits). Some reads are blocked unless the plan allows them (paid_plan_required and the other limit codes named in SKILL.md). Creating a monitor uses no credits, but each monitor alert uses 25 credits. If I have no Arcmira account, do not create one during this install: the API signup route (posting my email address to api.arcmira.com and posting back a six-digit code from my inbox) is a separate decision for later, never part of installing this skill file. Install "arcmira" for me. It is the shared procedure skill for Arcmira, the search engine for the spoken web: it teaches the agent to answer what YouTube shows and podcasts said — transcripts with timestamps, who was mentioned where, who sponsors whom, who recommends what on air (sponsored or organic), and whether talk about something is accelerating or fading. It works through the four tools of the remote Arcmira MCP server: arcmira_describe (returns the client reference: every method, its arguments and return fields, worked programs, quirks and error codes; call it once before the first program), arcmira_execute_read (runs one JavaScript program per question against the arcmira client and returns what the program returns; Premium transcripts included), arcmira_execute_write (the same client plus the monitor writes), and arcmira_feedback (tells Arcmira what went wrong). Its core rules: resolve a name to a verbatim id before using it as a filter (filters take ids only), pass as resolve context only words from my own message about that name, never a guess; say which entity was used, and say so when one was assumed from a "suggested" answer; when several rows fit and none stands out, bring the options back to me; build date windows from arcmira.today() and arcmira.daysAgo(n) (last 90 days when I name none); link each name to the page field the result carries, never build arcmira.com URLs by hand; cite returned passages and keep evidence from other sources distinct; an empty result means that query returned no matches, nothing more. When my ask matches one of the five task skills named in SKILL.md (sponsor-research, company-watch, find-quotes, person-research, compare-shows), SKILL.md says to load that skill and follow its worked program — those five skills are separate files, they are NOT in this folder and they are NOT installed by this install. Repository: https://github.com/arcmira/mcp (the skill is the whole plugins/arcmira/skills/arcmira/ 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/arcmira/ 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 844ea67207835136c74ae2af84d4578bfd75f30b445ce851b6274ff85d7ffbcd 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: - The skill folder holds 1 file only: SKILL.md (8,889 bytes), a pure instruction file (name: arcmira). There is no script, no code and no other file in the folder, and nothing in it can run by itself: every action it describes goes through the four tools of the external remote Arcmira MCP server (https://mcp.arcmira.com/mcp, backed by https://api.arcmira.com/v1/*). SKILL.md also points to five separate task skills (sponsor-research, company-watch, find-quotes, person-research, compare-shows) that live elsewhere in the same plugin (plugins/arcmira/skills/) and to Arcmira documentation pages (arcmira.com/docs, api.arcmira.com/v1/openapi.json, arcmira.com/llms.txt); none of that is part of this install, and this install does not connect the MCP server, does not sign me in, and does not create an account. - Nothing is sent anywhere during installation itself; the only download at install time is the skill folder from GitHub at the pinned commit. No script is run during installation (there is none), no key is asked for, no MCP server is added, and there is no telemetry in the folder. - 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 an Arcmira research question, and every send below goes to Arcmira's own servers (mcp.arcmira.com / api.arcmira.com) under my signed-in Arcmira account: (a) my question itself, sent as a JavaScript program to arcmira_execute_read or arcmira_execute_write: the program text carries my search query exactly as typed, the names, shows, topics and video ids or YouTube URLs I asked about, the date windows, the filters, and, for a resolve call, a context string that SKILL.md limits to words from my own message about that name. The arcmira_describe call sends at most a topic word. (b) My credential, handled by the host, never by the program: the OAuth token from my host sign-in (with the permissions I consented to), or my arc_sk_ account key as a Bearer header if I chose the key route; the repository README states the server-side sandbox adds the caller's credential to what it forwards, so the program text itself never holds the key. (c) Paid side effects inside a read: an arcmira_execute_read program that asks for a Premium transcript of a video that is not transcribed yet starts that transcription on Arcmira's side and is charged for it (the charge reports unit, amount and whether it came from included plan credits, on-demand budget, or both; a search charges 4 credits per passage past the first 5). (d) Writes, only through arcmira_execute_write and only after the monitor and its delivery are confirmed with me first, as SKILL.md requires: creating or updating a monitor sends the monitor name, the entities or exact names it follows (up to 90 ids per call), its notify_frequency (realtime, hourly or daily) and its delivery settings; a monitor then makes Arcmira send alerts about new mentions by email to my account address, or to Slack if I connect a Slack workspace (the skill reads my connected Slack workspaces and their default channel ids to set that up). Nothing is ever deleted by a monitor write; pausing is update(id, { paused: true }). Each alert uses 25 credits. (e) Feedback: SKILL.md ends by telling the agent to send one arcmira_feedback when anything was wrong, slow or missing for me, or when I want a source or capability Arcmira lacks — that call sends a category and a note saying what happened (and a call_id when a result named one) to Arcmira as a stored feedback record (POST /v1/feedback). It is a real send to Arcmira, it costs nothing, and it never happens during installation. (f) Logging by the MCP server, stated in the repository README section "What we log" (server behaviour, not code in this folder): each tool call is logged to Arcmira's product analytics (PostHog), linked to my account — one record per call holding the tool name, my host's name and version, the server version, the call latency, the input (the program text of either execute tool, first 4,000 characters; or the arcmira_describe topic; or the feedback category and note), the optional intent (at most 300 characters, my request in a few words) when the agent sends one, and the outcome only, not the result (ok or the error code, the result size, whether it was truncated, and the API routes called, without query strings). 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 local-only parts: the planning and the answer rules in SKILL.md (resolving names, checking which sense of a name a passage is about before asserting a mention, returning only the fields the answer needs, citing passages, naming the entity used) happen between you and me; the file itself reads no file on this machine, reads no environment variable and stores nothing. - Do not add the Arcmira MCP server, do not sign me in, do not ask for or set any Arcmira key, do not create an Arcmira account, do not call arcmira_describe, arcmira_execute_read, arcmira_execute_write or arcmira_feedback, and do not run any research, transcription or monitor during this install: connecting the service and spending credits are my decisions, later, when I actually ask an Arcmira question. 3. When I say "yes, install", install it as a skill: copy the whole arcmira folder into the agent's skills directory, in a folder named "arcmira", keeping its 1 file (SKILL.md). A manual clone followed by copying the plugins/arcmira/skills/arcmira 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: arcmira), 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: 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 through the host with my Arcmira account — or set an arc_sk_ account key as a Bearer header if my client cannot sign in — knowing that metered reads spend my plan credits first and then my dashboard on-demand budget, that a Premium transcript of an untranscribed video starts a charged transcription, that monitor alerts cost 25 credits each and are delivered by email to my account address or to a Slack workspace I connect, that each tool call is logged by Arcmira to its product analytics as disclosed in point 2, and that the five task skills SKILL.md points to are separate installs; using the skill is a separate step that only starts when I ask for it, and the use-time sends and charges disclosed above 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.