TaskTreker publishes step-by-step guides for the parts that are easy to get wrong: the first week with the assistant, connecting capture sources, keeping a live sizing document, preparing a handoff for sales, building a harness of your own prompts and connecting an MCP server. Every guide has a goal, a step-by-step path, the result you should get, and a section on what to do when it does not work.
Step-by-step instructions for each part of the product.
A step-by-step plan for the first seven days: what to set up on day one, what to leave alone for now, and how to tell that the assistant is working.
What you get: By the end of the week you have a confirmed context, a working capture loop on at least one real channel, a cleared confirmation queue and a weekly ritual you can repeat.
How to choose capture channels, set the rights for each one, add material by hand and use the universal import path when your tools are not connected.
What you get: A capture setup with a defined channel list, explicit rights per channel, a working import path, a consent log you can show your team and a verification step that proves the setup reads what you think it reads.
How to keep market sizing inside the product as a set of sourced assumptions with cascading recalculation and dated versions, instead of a file that goes quietly stale.
What you get: A sizing document where every number traces back to a sourced assumption, a change to one assumption recalculates the dependent figures, and a dated version with the full assumption list is ready to present.
How to turn the accumulated context into a package of nine sourced artifacts with a confirmation gate and a delivery channel, at the moment a feature moves to the release stage.
What you get: A confirmed package of nine artifact types, each carrying its own sources and gaps, delivered to the channels you chose, plus a task draft for the tracker in the format you picked, with acceptance criteria.
How to set up the harness layer of the assistant: read the prompt library, put rituals on a schedule, version your edits with an evaluation set, and read the quality dashboard to see where the assistant was wrong.
What you get: A configured harness: a prompt library where every prompt shows its instruction, input, output and quality criterion; a version history with an evaluation score and a one-click rollback; day, week and month rituals on a schedule; and a quality dashboard with take rate, evaluation score and recoverability.
How to connect the TaskTreker MCP server to a client agent, keep the access read-only by default, and get answers about your own product context with citations to the sources.
What you get: A connected MCP server: the client agent sees the context of your company and nothing else, answers questions with citations and freshness marks, and the default access mode stays read-only with every call visible in the audit log.