Projects
The memory and seatbelt for your AI coding agent.
An agent can’t write code until a person approves the approach, the spec passes a completeness check and a failing test is on record. Every change lands on an audit ledger.
You are the agent: take the proper step, or try a shortcut.
Try the seatbelt
You are the agent. Story: checkout rejects an empty cart. Take the proper step, or try a shortcut and watch the gate stop it.
- Brainstormapproach-approval
- Speccompleteness
- Testred-before-green
- Developspec-scope
- Reviewspec-then-quality
- Shiphuman-landed
Next step
Brainstorm is open. The agent cannot write implementation until a person approves exactly one approach.
Run-state gates on the agent’s own file writes
- approach-approvalblocking
- spec-persistedblocking
- no-code-without-failing-testblocking
- spec-scopewaits for spec
$ mokata init --profile standard --yes
mokata initialized with profile ‘standard’.
$ /brainstorm "checkout rejects an empty cart"
mokata · brainstorm (engaged) two approaches on the table:
A raise ValueError on an empty cart
B return 0 and log a warning
The spec-persisted, failing-test and memory-write messages are mokata’s published output. The other lines are paraphrased from its docs.
- MCP tools
- 62
- Agent skills
- 26
- Enforced gates
- 9