One Slack Agent the Whole Team Asks About Go-to-Market

To give the whole team one place to ask about go-to-market, deploy Cargo’s ask-cargo cookbook: one Claude Code agent people @mention in Slack, which answers from your go-to-market repo and the live workspace with a source for every answer, turns change requests into pull requests, and runs anything that spends only after someone in the thread replies with a go. It never deploys, merges or deletes.

What the cookbook deploys, from its code in getcargohq/gtm-skills on 2026-10-16: one agent with a Slack trigger on every channel the bot is in, three connectors, a folder and a viewer service user. It runs on no schedule, and a batch over 25 records runs a sample of 5 first.

The problem, and the answer #

Questions about the go-to-market stack (“what is our ICP”, “did the scoring play run”, “add this competitor”) end up with whoever wrote the code, and the answers live in three places: the repository, the workspace’s run history and someone’s head. The answer is an agent with a checkout of the repository and read access to the workspace, sitting where the team already asks.

How it works, end to end #

1. Someone @mentions the bot in a channel it has been invited to. The trigger opens one conversation per Slack thread and marks the thread as working.
2. The agent clones your repository, resolved from your checkout’s git origin at deploy, and reads before it answers: context/ (ICP, personas, competitors, objections), cadence/ (daily logs, weekly plans), initiatives/, infra/ for what is actually declared, and git log.
3. It reads the workspace with Cargo’s CLI, signed in as the ask_cargo service user with the workspace’s viewer role: cargo-ai orchestration run list, batch list, cargo-ai billing usage get-metrics, cargo-ai ai agent list, cargo-ai storage model list, a sampled storage query execute, and cargo-ai project check or project plan for what a change would do.
4. It replies in the thread, citing the file path, command or pull request behind the answer. The platform streams the reply; the agent posts nothing itself.
5. A follow-up @mention in the same thread lands in the same conversation, so the agent sees its own earlier reply and any go together.

What it does besides answering #

• Changes become pull requests. “Add this competitor” becomes an edit on a branch named ask-cargo/<request> and a pull request linked in the reply. One thread is one pull request; a follow-up adds a commit.
• Anything that spends waits for a go. Executing an action, creating a batch, running a play, messaging another agent or writing to a CRM gets a proposal first: what will run, over how many records, and the total in credits from the live catalog. It runs only after a later message in the thread approves that proposal, and a go never covers more records or more cost than proposed. Over 25 records, it runs 5, replies with the result, and asks for a second go.
• Work goes to the agent that owns it. When a deployed agent already owns the job (a standup, a weekly planner, an account scorer), it proposes the handoff and, on a go, sends it with cargo-ai ai message create instead of redoing the work.
• A thread worth keeping is captured. “@Cargo capture this” writes the thread verbatim to cadence/log/raw/slack/, an entry with objections, competitors, signals, product asks and actions to cadence/log/slack/, and a context/ update only when a claim now has two independent occurrences across calls and threads. It reads only that thread, and refuses shared channels and direct messages.

What it needs and what it costs #

• A go-to-market repository, the shape cargo-ai project init scaffolds.
• GitHub, Slack and Anthropic connectors. The Slack connector is the trigger and the reply path.
• Where the bot is. The trigger answers in every channel the bot has been invited to, so keep it out of customer shared channels, or list channels explicitly instead.
• @cargo-ai/cdk 1.0.104 or later, which brings the service user the CLI runs as.

Every turn is an agent run: a sandbox, a clone of the repository and language model tokens billed through your Anthropic connector. A turn that only answers costs tokens and nothing else, because the read-only cargo-ai calls are API reads that bill nothing. What spends is only what someone approves with a go, priced in the proposal before it runs.

What it will not do #

• Deploy, merge, destroy or remove anything, with or without a go. Its service user holds the viewer role, which cannot deploy or change a resource even if the prompt were ignored.
• Mint or rotate tokens, change workspace members, or touch billing.
• Post anywhere but the thread it was asked in, or answer direct messages.
• Run anything that spends without a go in the thread, or send to people the workspace has no consent basis for.
• Know who wrote a message. Anyone in the thread can steer or approve, so pull requests quote the request and name no one.

Only @mentions wake it, follow-ups included, and a mention sent while it is still working is dropped, which is why every reply that needs a confirmation says exactly what to type. If the team only needs answers from context and models, the workspace’s built-in Master Agent with a Slack trigger covers that without a checkout.

Deploy it #

npm install -g @cargo-ai/cli
cargo-ai login --email you@company.com
cargo-ai project add cookbook/ask-cargo
cargo-ai project plan
cargo-ai project deploy

Signing up from the CLI needs no browser and no card, and comes with 100 free credits. Fill references/roster.md with the agents your workspace runs (cargo-ai ai agent list) so handoffs go to the right one. The ask-cargo cookbook page lists every resource and placeholder.

It hands off to the daily go-to-market standup and weekly plan when those are deployed, and its captures land beside every sales call in your go-to-market repo.

Frequently asked questions #