Most tools generate docs. ZeNorm generates pushback.
Every spec below started as one vague sentence. The agent read the codebase, asked hard questions, and the server refused to emit a single task until the intent, constraints, and acceptance criteria each cleared a 0.9 quality bar. Nothing here is edited or staged — these are complete records of real runs, refusals included.
- Activity feed and audit log5 refused
Add truthful attribution, workspace-wide activity feed, and append-only audit trail
Agent edits are recorded as if the human typed them. Asked what immutability can honestly mean here, the agent declines to promise tamper-evidence it cannot deliver.
“When something changes in a workspace nobody can tell who did it or when.”
4 rounds1 questions16 blocks18 trace edges - Notifications for spec and task events5 refused
Add org-scoped notification center and email digests for spec activity
The agent audits the existing activity spine, finds five event types declared but never written, and the PM refuses to build notifications on a spine that cannot report an assignment.
“We need to tell people when things happen on their specs instead of making them check the board.”
5 rounds5 questions18 blocks22 trace edges - Public REST API + outbound webhooks5 refused
Org-scoped API keys, spec read API, and verifiable outbound webhooks
Asked whether anything in the schema would be a mistake to publish as a permanent contract, the agent finds two untyped JSON blobs and the answer becomes a constraint.
“Teams want to hook ZeNorm into their own tools.”
4 rounds2 questions17 blocks24 trace edges - Spec templates and reusable scaffolds4 refused
Spec templates: author, promote, manage, and create from scaffolds
A template that pre-fills goals could let a spec pass the quality gate on boilerplate nobody thought about — the agent traces the scoring code and proves which design avoids it.
“Every spec starts from a blank page and people keep rewriting the same things.”
5 rounds2 questions17 blocks24 trace edges - Usage metering and budget enforcement1 refused
Usage visibility dashboard and org-settable budget gates for agent spend
The PM argues against their own instinct on concurrent spend, and the agent lands on a cheaper mechanism that bounds the overshoot to a number the product can promise.
“I have no idea what our agent runs actually cost us or how to stop a runaway bill.”
6 rounds3 questions16 blocks26 trace edges
A real request, the questions the agent asked, and every place the server refused its output. No product announcements.