CompetLab Agent Skills

What they are

An Agent Skill is a playbook an AI coding agent loads and follows — a SKILL.md file (following the agentskills.io  open standard) that describes a task, when to use it, and how to do it well. CompetLab Agent Skills package the work of reading competitive-intelligence data correctly: each skill knows which CompetLab tools answer its question, how to read what comes back, and how to shape the result into a deliverable you’d hand to a founder, a PMM, or a sales team.

CompetLab monitors your competitors across six dimensions — AI Visibility, AI Sources, Positioning, Pricing Intelligence, Content Intelligence, and Tech & Trust Profile — and writes a Strategic Briefing over 14 analysis areas. The skills do not do the research. The platform does, with more sources than an agent can reach in a session and a memory of prior readings. The skills make sure what comes out the other end is true.

The MCP server gives your agent the raw tools; Agent Skills give it the playbook. On its own, an agent with MCP access can read your dimensions and competitors, but you still have to tell it what a good battlecard looks like — and an agent given competitive data will happily round, pool, and order it into something confident and wrong. A skill carries the method and the reporting rules with it, so “battlecard for Rival Inc” produces a structured, sales-ready card instead of an improvised summary.

The suite

Seven skills. Every one answers a single question and reads the platform to answer it.

Every skill, with what it reads and the phrasings that trigger it, is in the skill reference.

How they work

A skill reads your data through the CompetLab MCP server — it calls the mcp__competlab__* tools, the same ones documented in the MCP tool reference. The briefing, dossier, and battlecard skills read the platform’s Strategic Briefing for the researched areas — funding, hiring, product launches, reliability, customer voice, developer ecosystem, agent adoption — rather than re-researching them with a web search that would return a thinner answer with nothing to compare it against. The output is always plain Markdown — a briefing, a dossier, a card, a work list — that you can read, edit, and share.

Because the skills run on the MCP server, they see exactly what your API key can see and nothing more.

Every skill ships with the reporting discipline the platform enforces on itself, as a reference file the agent reads before it writes a sentence:

• null means not measured. Never zero, never empty, never “no”. A real 0 is a finding; a null is a gap in the reading, and the two never share a sentence. One carve-out: a rank is null on a brand named in no answer, and that is measured — not named in any answer.
• A page is not the whole list. The long lists arrive one page at a time; the count is the page object’s total, never the rows the agent holds.
• Counts, never rates. “Named in 8 of 69 answers”, not “12%”. The question set is small by design, and a share computed from it is false precision.
• Per engine, never pooled. The engines read different pages; a combined figure describes a list none of them produced.
• Retrieved, never cited. The engines do not disclose which pages they leaned on, so a citation count is a number that does not exist.
• Overlapping ranges are not ordered. Two brands whose ranges overlap are tied, and a difference between two overlapping readings is two readings, not a movement. Order is by presence — how often a brand is named — and Google AI Overviews names companies in prose and ranks nothing.

What you need

• A CompetLab account with at least one project and its competitors being monitored. One exception: competlab-site-audit runs on any public domain with no project and no competitors configured — a free-trial key is enough.
• The CompetLab MCP server configured with your API key — this is how every skill reaches your data, the site audit included. See the Connect guide to set it up. A read key is enough for the whole suite.
• A compatible agent. The skills are written for Claude Code (the plugin marketplace and the mcp__competlab__* tool names are Claude Code conventions). Because they’re standard SKILL.md files, Cursor, Codex, Gemini CLI, and any agent that reads the format can use them too — you may just need to adjust the MCP tool names for your client.

Authentication lives in the MCP configuration, not the skills — there’s no separate skill-level API key. The Setup guide walks the whole thing.

What they don’t do

A few honest limits:

• They read your account; they don’t change it. Skills pull your monitoring data and run the public scans; nothing a skill does writes to or changes your CompetLab projects, competitors, prompts, alerts, or schedules. competlab-monitoring-setup recommends changes; a person makes them in the app.
• They don’t re-research what the platform already researched. Competitor funding, hiring, launches, reliability, reviews, and developer ecosystem are sections of the Strategic Briefing. The skills read those sections; they don’t go and fetch them again.
• They don’t hand over a generic checklist. competlab-ai-sources shows what the engines read before answering; it doesn’t promise that working the list produces recommendations, because the factors differ by category and training data often outweighs what the models read.
• A site audit doesn’t predict what the models say. competlab-site-audit reports what an automated reader can extract from a site. What AI models say about the brand is competlab-ai-visibility, and nothing in the audit predicts it.
• Every dimension check and briefing edition costs real money to produce. The skills read what the platform has already produced before asking it to produce more, and they never trigger a scan in a loop.

Next steps

• Setup → — install the plugin, connect the MCP server, and run your first skill.
• Skill reference → — all seven skills with what each reads and when to use it.