GrowthMap

Work with agents

A conversation becomes a map you can correct, continue, and connect to outcomes.

For a Personal Finance App, the person owns direction while the Agent records confirmed knowledge.

01 / A complete user story

1. Discuss the goal

Say: “I want a Personal Finance App with no bank connection and three-second entry.” The Agent clarifies audience, limits, and success.

2. One confirmed item, one node

The Agent records a goal, feature, decision, task, and risk separately—not as one chat summary.

3. Human correction

Rename, move, delete, or rewrite anything. Guesses stay off the Mainline; people decide direction.

4. Continue next time

The next conversation starts from the current map instead of repeating confirmed context.

5. Read outcomes back

After Quick entry is implemented, a person or authorized Agent records summary, files, commit/PR, tests, risks, and next steps on that node.

02 / Collaboration boundaries

Limited write, in plain language

Agents do not open SQLite, the filesystem, or a shell directly. Reads and writes pass through the localhost application API. Grants scope project, node, branch, time, and operations; revisions prevent stale overwrite; receipts record accepted operations.

No automatic Agent execution

GrowthMap organizes context and bounded writes. It does not run Agents or automatically scan Git, filesystems, or deployments.

No named one-click integration claim

Compatible tools must use the local interface themselves; no named Agent currently has a certified one-click setup.

03 / Developer entry

Protocol details live in Developer docs

REST/JSON, CLI/MCP, grants, revisions, receipts, Context Packets, and integration limits are optional advanced topics.

Open Developer docs