| name | policyvault-repo-workflow | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| workspace |
|
||||||||
| agent |
|
||||||||
| approval_policy | on-request | ||||||||
| sandbox | workspaceWrite | ||||||||
| network_access | restricted | ||||||||
| policies |
|
||||||||
| hooks |
|
This file exists for two reasons.
First, the repo should be easy to operate inside the Codex app. Second, the project itself is supposed to demonstrate disciplined AI-native execution, not just fast code generation.
When an agent works on this repo, it should:
- respect the architecture and file boundaries
- keep work inside the assigned submilestone
- update the active ExecPlan while working
- run validation before proposing completion
- preserve a clean interview story
Unless an ExecPlan says otherwise, a code task in this repo is only complete when:
pnpm compilepassespnpm testpasses- touched docs are updated
- the active ExecPlan is updated
- new state changes are covered by event-aware tests or explicit event assertions are added later in the same milestone
- any UI contract changes are reflected in the ABI/address sync notes
This repo is intentionally smaller than Pocket CTO. The workflow contract still exists because it helps control Codex scope and keeps the implementation explainable.
For non-trivial work, start the thread with a skill bundle from .agents/skills/README.md so the Codex operating pattern is explicit instead of implied.