> ## Documentation Index > Fetch the complete documentation index at: https://docs.nephia.cc/llms.txt > Use this file to discover all available pages before exploring further. # Skills > Agent recipes for Nephia as SKILL.md files, from a daily brief to churn signals, installed beside the MCP server and run on your own schedule. A skill is a markdown file an agent reads before a job: when to run it, which Nephia tools to call, in what order, and what it must not do. Six ship in the [`nephia-mcp`](/mcp-server) package, under `skills/`. They are the same recipes as the server's [prompts](/mcp-server#prompts), written for an agent that runs them unattended. They need the [MCP server](/mcp-server) connected: a skill calls tools, it holds no data of its own. ## Install With the [skills CLI](https://github.com/vercel-labs/skills), from the package: ```bash theme={null} npm install nephia-mcp npx skills add ./node_modules/nephia-mcp/skills ``` Or copy one folder, for example `skills/daily-brief`, into your agent's skills directory (`~/.claude/skills/` for Claude Code). Fill the placeholders, `{{brand}}` or `{{slack_channel}}`, in your copy. ## Run it on a schedule Nephia keeps no schedule for a skill. The recurrence lives with the agent: a scheduled task in Claude or ChatGPT, a cron job running Claude Code, or an [n8n](/integrations/n8n) workflow. Nephia keeps collecting in between, so a skill that runs at 8:00 reads everything caught overnight. ## What a skill never does Every skill reads and reports. None of them answers anyone on X, Reddit or any other platform, and none of them speaks for you in public. Delivering a brief to Slack or opening an issue in Linear goes through **your** agent's own connector: Nephia never connects to those tools. A skill that can spend, like a weekly digest summarised by AI, prices the call first with the matching estimate tool, shows you the figure, and runs it only after you agree. No skill quotes a price: the figure is in each tool's own description, which is where it stays correct. A test in the package refuses any skill that names a tool the server does not have, types a credit figure, or tells the agent to act on your behalf in public. ## The six skills ### alternative-seekers `packages/mcp/skills/alternative-seekers/SKILL.md`, installed as `skills/alternative-seekers/SKILL.md`. ```markdown theme={null} --- name: alternative-seekers description: Build a reading list of threads where someone looks for an alternative to {{competitors}} or compares tools, ranked by engagement, from what Nephia caught. --- # Alternative seekers ## When to use Weekly, for a founder or a sales team that wants to know where buyers are comparing options right now. ## Tools - `list_mentions`: the comparison and purchase-intent stream, and a free text search for each competitor. Every call here is free in the default text mode. ## Steps 1. Call `list_mentions` with `intent: ["comparison", "purchase_intent"]` and `since` set to 30 days ago. 2. For each name in {{competitors}}, call `list_mentions` with `q` set to that name, then once with `q` set to `alternative`. A text search reads the window in slices: a page with no mentions that still carries `nextCursor` is not the end, so follow `nextCursor` until it is absent before concluding nobody said it. 3. Keep threads where someone is choosing: asking for an alternative, comparing tools, unhappy with a competitor, about to buy. Drop reviews, news and vendor announcements. 4. Rank by engagement, then by recency. 5. For each thread: link, Source, author, one quoted line, the competitor it is about, and why it qualifies (`intent`, `matchedTerms`). ## What this skill does not do This is a reading list. It drafts no answer and contacts nobody: the person decides which conversations to join, and joins them themselves. ``` ### churn-signals `packages/mcp/skills/churn-signals/SKILL.md`, installed as `skills/churn-signals/SKILL.md`. ```markdown theme={null} --- name: churn-signals description: Spot customers of {{brand}} who may be leaving, from Nephia mentions that say cancel, switched or alternative, and from authors who complain again and again. --- # Churn signals ## When to use Weekly, for a customer success or founder team that wants to hear about unhappy customers before the cancellation. ## Tools - `list_mentions`: negative mentions, text search with `q`, and the `author` filter to read one person's history. Every call here is free in the default text mode. Its description states what `mode: "semantic"` would cost; this skill does not need it. ## Steps 1. Call `list_mentions` with `sentiment: ["negative"]` and `since` set to 30 days ago. Page with `nextCursor` until the window is read. 2. Call it again with `q` set to each of: `cancel`, `switched`, `alternative`, `refund`, `leaving`. Follow `nextCursor` until it is absent: a text search reads the window in slices, and an empty page that still carries a cursor is not the end. Keep what mentions {{brand}} itself rather than a competitor. 3. Count mentions per author. For every author with two or more negative mentions, call `list_mentions` with `author` set to that handle and `since` 90 days ago, to see whether the tone is getting worse. 4. Rank the signals: an explicit intent to cancel or a completed switch first, then repeated complaints from one author, then single complaints with high engagement. 5. For each signal: the author, the Source, the quotes with dates and links, what they are unhappy about in one sentence, and the owner from {{account_owner_map}} if the person gave one. ## What this skill does not do It contacts nobody, answers nobody in public and changes no Query. The team decides who to reach out to, through their own channels. ``` ### daily-brief `packages/mcp/skills/daily-brief/SKILL.md`, installed as `skills/daily-brief/SKILL.md`. ```markdown theme={null} --- name: daily-brief description: Every morning, read what Nephia caught in the last 24 hours and deliver the five mentions of {{brand}} to deal with first, each with why it is on the list, after checking that every Source was actually checked. --- # Daily brief ## When to use Run it once a day, from a scheduled task in your agent, a cron job or an n8n workflow. Nephia keeps the mentions; the schedule is yours. ## Tools - `list_queries`: every Query, with each Source's health and `staleSources`. - `list_mentions`: the stream, filtered by sentiment and intent. - `account_credits`: the balance, for the last line. Every call in this skill is free: leave `mode` at its default text mode. Each tool's description states what it costs. ## Steps 1. Call `list_queries`. Note every Source whose `health.stale` is true or whose `health.lastOutcome` is `"failed"`, with its Query and `lastCheckedAt`. A Source nobody checked says nothing about what was said there, so these go first in the brief. 2. Call `list_mentions` three times with `since` set to 24 hours ago: `sentiment: ["negative"]`, `sentiment: ["question"]`, and `intent: ["purchase_intent", "comparison"]`. Follow `nextCursor` only while the page is inside the window. 3. Pick five, in this order of priority: someone choosing or comparing tools, a question nobody answered, a complaint, then the most engagement. Keep one mention per thread. 4. Write one short block per mention: Source and author, one line quoted exactly, why it is there (`matchedTerms`, sentiment, intent, engagement), the link, and the next step in one sentence. 5. Close with the late or failing Sources and the balance from `account_credits`. 6. Deliver the brief to {{slack_channel}} through your own Slack connector, or show it in the conversation. ## What this skill does not do It answers nobody on X, Reddit or anywhere else, and it changes no Query. The person reads the brief and decides. ``` ### mentions-to-issues `packages/mcp/skills/mentions-to-issues/SKILL.md`, installed as `skills/mentions-to-issues/SKILL.md`. ```markdown theme={null} --- name: mentions-to-issues description: Turn bug reports and feature requests about {{brand}} found by Nephia into issues in {{issue_tracker}} (Linear or GitHub), each linking back to the original mention, without duplicates. --- # Mentions to issues ## When to use Daily or weekly, for a product team that wants what users report in public to land in its tracker. ## Tools - `list_mentions`: complaints and questions from the window. - `get_query_results` and `list_query_buckets`: when a Query sorts mentions into buckets such as "Bugs", read that bucket directly. All of these are free in the default text mode. The issue tracker is **your** connector (Linear, GitHub): Nephia never connects to it. ## Steps 1. Call `list_mentions` with `intent: ["complaint", "question"]` and `since` set to the last run of this skill (24 hours ago by default). 2. If a Query sorts mentions into a bug or feature bucket, call `list_query_buckets` for its id and read that bucket with `get_query_results` and `bucket` set to the bucket id. 3. Keep only mentions that describe a concrete defect or a concrete request. Drop general praise, pricing opinions and anything you cannot reproduce in one sentence. 4. Search {{issue_tracker}} for an existing issue on the same problem. If one exists, add the mention's link to it as a comment; otherwise create one. 5. Each new issue gets a title in the product's words, the quote verbatim, the author, the Source, the date, the link to the mention, and the label {{label}}. 6. Report what you created and what you attached to existing issues. ## What this skill does not do It answers nobody on the platform where the mention was found, closes no issue and changes no Query. ``` ### wall-of-love `packages/mcp/skills/wall-of-love/SKILL.md`, installed as `skills/wall-of-love/SKILL.md`. ```markdown theme={null} --- name: wall-of-love description: Collect specific, positive quotes about {{brand}} from Nephia, each verbatim with author, Source, date and link, grouped by theme, ready for the person to request permission to reuse. --- # Wall of love ## When to use Before a launch, a landing page update or a sales deck: when you need proof in customers' own words. ## Tools - `list_mentions`: positive mentions and praise. - `analyses_estimate` and `analyses_run`: an optional AI sort into themes for a long list. `analyses_run` charges; its description gives the rate. ## Steps 1. Call `list_mentions` with `sentiment: ["positive"]` and `since` set to 90 days ago, then again with `intent: ["praise"]`. Follow `nextCursor` until the window is read, merge, and keep one mention per thread. 2. Keep quotes that say what the product did, in the author's own words. Drop sarcasm, passing mentions and anything from an exchange with support. 3. Group them by theme yourself. Only for a long list, price a sort with `analyses_estimate`, show the figure, and call `analyses_run` with `kind: "classify"` once the person agrees. 4. Return each quote verbatim with author, Source, date and link, strongest first, and save the list to {{destination}} if the person named one. ## What this skill does not do A quote belongs to its author. This skill contacts nobody and puts nothing on a public page: the person obtains permission first. ``` ### weekly-digest `packages/mcp/skills/weekly-digest/SKILL.md`, installed as `skills/weekly-digest/SKILL.md`. ```markdown theme={null} --- name: weekly-digest description: Once a week, summarise what people said about {{brand}} in Nephia, with volume by Source, the sentiment split, three themes and five verbatim quotes with links. --- # Weekly digest ## When to use Monday morning, for a team that does not read the stream every day. ## Tools - `list_queries`: Source health, to say which Sources were late. - `list_mentions`: the week's mentions, paged with `nextCursor`. - `analyses_estimate` and `analyses_run`: an optional AI summary over a window too large to read. `analyses_run` charges; its description gives the rate. ## Steps 1. Call `list_queries` and note every Source with `health.stale` set to true. 2. Page `list_mentions` over the last 7 days (`since` 7 days ago, the default page size, `nextCursor` passed back until it is absent). Tally by Source and by sentiment, and say how many mentions you read. 3. Name three themes that keep coming up, each with one line of explanation and a count. Read them yourself. Only if the window is too large to read, price a summary with `analyses_estimate`, show the person the figure, and call `analyses_run` with `kind: "summarise"` only once they agree. 4. Choose five quotes, verbatim, each with author, Source, date and link. 5. Lay it out: volume by Source, sentiment split, three themes, five quotes, late Sources. Deliver it to {{slack_channel}} through your own connector, or show it in the conversation. ## What this skill does not do It changes no Query, answers nobody, and runs no paid analysis without the person agreeing to its price. ```