Company Watch

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 — an external MCP service and an Arcmira account are required to USE this skill, though not to install its file: this skill is one Markdown file with no local code, and it does nothing on its own. It works only in an agent whose host is connected to Arcmira's remote MCP server at https://mcp.arcmira.com/mcp (which fronts https://api.arcmira.com/v1/ under my account) and signed in to an Arcmira account: normally the host signs me in through OAuth (the server answers 401, the host registers against api.arcmira.com and opens arcmira.com for sign-in and consent, and the token carries only the permissions I allowed); a client that cannot do OAuth instead sends an Arcmira account key (arc_sk_...) as a Bearer header in its own MCP configuration. Reads are metered in credits from my account's plan, then its on-demand budget (see point 2: a momentum call uses 10 credits, a search uses 4 credits per passage returned past the first 5, creating a monitor uses no credits, and each monitor alert uses 25 credits). Do NOT connect the MCP server, sign me in, ask me for an account key, or configure anything during this install: connecting Arcmira is my own separate step, before or after, and this install only copies the skill file. Install "company-watch" for me. It watches a company, product, brand, person or topic on the podcasts and YouTube shows Arcmira indexes: it answers "what was said about X lately" in one read program over the last 90 days unless I name another window — episode counts per show from occurrences, the rising-or-fading verdict from momentum, the catalog notes from mentions, and quotes from search — then, only if I ask to keep following it, it finds a monitor that fits (reading my existing monitors and my connected Slack workspaces first) or, after I confirm the monitor and its delivery, saves a new one in a separate write step. Names are never used as filters directly: the program first resolves the exact name I said (a company with no type, a person as a person, and a topic under every spelling I mean, for example "data centers", "datacenters", "data centre", "data center", keeping every distinct id); a suggested pick is an assumption and the answer must name it with its evidence, and when several entities fit, the options come back to me and nothing else runs. Repository: https://github.com/arcmira/mcp (the skill is the whole plugins/arcmira/skills/company-watch/ 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/company-watch/ 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 74d5a7ba13f407bc8a3336372e0f79d4fcf5e0891aba2df99f8b30027affa877 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 (12,131 bytes). There is no script, no executable code, no package and nothing that runs locally in the folder: SKILL.md is instructions plus three worked JavaScript programs that the agent passes to Arcmira's remote MCP tools (arcmira_execute_read for the two read programs, arcmira_execute_write for the save program; arcmira_describe holds the method reference). The folder is not functionally self-contained: it needs the external Arcmira MCP server named in the prerequisites, and the plugin manifests that declare that server (plugins/arcmira/.mcp.json and plugins/arcmira/mcp.json in the repository) are NOT part of this install — installing this skill does not connect, configure or sign in to anything. - 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 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 what was said about a subject or ask to follow one: (a) Research turn, via arcmira_execute_read to https://mcp.arcmira.com/mcp: the agent sends the whole JavaScript program as text, and the program carries the exact name I said plus my own words about it as context for resolve (never a guessed context), then the resolved entity id, my date window (90 days by default) and the call limits, to momentum, occurrences, mentions and search (the search query is the entity's own name); the server runs the program against api.arcmira.com under my account credential, which its outbound proxy adds — the program itself never holds my key or token. (b) Monitor-discovery turn, also arcmira_execute_read: the program resolves every name and every topic spelling I gave, and reads back, under my account, my own Arcmira monitors (monitors.list and what each of the first 10 follows, via monitors.trackers) and my connected Slack workspaces (integrations.slack: workspace id, team name, default channel id and channel name) so an existing monitor and delivery can be suggested. (c) Save turn, via arcmira_execute_write, ONLY after I confirm the monitor and its delivery: it creates a monitor on my Arcmira account (name and notify_frequency: realtime, hourly or daily; plus notify_slack with the slack_integration_id and slack_channel_id from step (b) if I chose Slack) or updates an existing one (delivery change, or paused: true to pause; nothing is ever deleted), then follows the ids I agreed to with addEntities (up to 90 in one call) and any exact name resolve did not find with addName (name plus type). Afterwards Arcmira itself — not this machine — sends the alerts for what the monitor follows, by email to my account address or to that Slack channel, at the frequency I confirmed. (d) Feedback, after an answer: SKILL.md says that if anything was wrong, slow or missing, or I want a source or capability Arcmira lacks, the agent sends one arcmira_feedback: a category and a note describing what happened go to Arcmira (one POST /v1/feedback; the repository documentation adds that an optional request_id or call_id may name the call). No feedback is sent during installation. - Credits and gates at use time, from SKILL.md and the repository documentation: metered reads draw credits from my plan first, then the on-demand budget I set in the Arcmira dashboard — momentum uses 10 credits a call, search uses 4 credits per passage it returns past the first 5, creating a monitor uses no credits, and each monitor alert uses 25 credits; a read the plan or budget blocks comes back as an error naming the limit (paid_plan_required, quota_exceeded, spend_limit_exceeded and the like); a sign-in that lacks the monitors:write or trackers:write scope fails the save with insufficient_scope until I reconnect Arcmira allowing monitor changes. - Call logging, from the repository README section "What we log" (outside the skill folder, read when this listing was prepared): each Arcmira tool call is logged to Arcmira's product analytics (PostHog), linked to my account — tool name, host name and version, server version, duration, the input (the program text of an execute call, first 4,000 characters; or the feedback category and note), an optional intent of at most 300 characters when the agent sends one (my request in a few words), and the outcome, not the result: ok or the error code, result size, whether it was truncated, and the API routes the call made (the route, not its query string). Before storage, 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. - Honest scope, from SKILL.md: the counts measure only the shows Arcmira indexes, not the internet — an empty window means no indexed show said it, and the answer must cite the index date (as_of) before saying nothing happened; episode counts come from occurrences, never from counting mentions rows; speaker attribution and catalog notes can be wrong (a title far from the company is a homonym the catalog mislabelled), so titles and notes are read before asserting a lone mention. - Do not connect the Arcmira MCP server, do not sign in, do not ask for or handle an account key, do not run any of the three programs, do not call arcmira_describe or arcmira_feedback, and do not create, update or pause any monitor during this install: all of that is my decision, later, when I actually ask for a watch. 3. When I say "yes, install", install it as a skill: copy the whole company-watch folder into the agent's skills directory, in a folder named "company-watch", keeping its 1 file (SKILL.md at its root). A manual clone followed by copying the plugins/arcmira/skills/company-watch 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: company-watch), all 1 files in place with matching fingerprints. Do not run any program, and do not contact the Arcmira MCP server, to verify. 5. Report what was installed, where, and what I still need to do myself (for example: connect the Arcmira remote MCP server https://mcp.arcmira.com/mcp in my host myself and sign in through OAuth, or put my own Arcmira account key in my MCP configuration myself — never hand it to the agent in this install; expect metered reads and monitor alerts to draw credits from my plan, then my on-demand budget; give the exact name and, for a topic, its spellings, and pick myself when resolve returns several options; confirm the monitor and its delivery — email to my account address or a connected Slack workspace — myself before any save, and pause a monitor myself later if I want it stopped; using the skill is a separate step that only starts when I ask for it, and the use-time sends disclosed above happen only then, on the turns that run). 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.