How to use Codex to build document-backed household tasks
Codex built ClearDay’s source-grounded task prototype with visible document conflicts, persistent evidence and drafts; fictional samples imply no real actions.
devpost.com
Suggested prompt
Suggested prompt written by Agents In Use, not the source author. Replace bracketed inputs and review tool permissions first.
Build a life-admin prototype from these authorized sample documents. Keep conflicting dates visible, verify evidence excerpts against text, label live and fallback analysis, and scope saved tasks to the authenticated user. Preserve drafts and downloads as review checkpoints; do not send, pay or submit, and document incomplete account and deletion flows.
Reach people exploring AI agents
See placements →Evidence at a glance
Sourced account- Source author
- Nithish K
- Original account
- Read the source ↗
- Published
- Source link checked
- Reported result
- The creator reports a working prototype with persistent evidence-backed tasks and disclosed fallback behavior; fictional samples do not establish completed real household administration.
- Human involvement
- The owner defined privacy and approval boundaries; users review evidence and conflicts, and the app does not send, pay, sign or submit automatically.
A linked account supports attribution. It does not mean we independently reproduced its result.
Report an issue with this case
Open correction emailOpens your email app with the case link included. Nothing is sent automatically.
How the agent did it
- Nithish K defined a source-grounded, approval-first prototype and supplied fictional conflict samples.
- Codex implemented validated analysis, user-scoped persistence and separately labeled live and fallback modes.
- The workflow retained human source review and draft approval, with external sending and payments outside the implemented actions.
Nithish K used one primary Codex session to build ClearDay, a working life-admin prototype that turns source documents into reviewable tasks. Its sample pack includes a lease notice, school consent details and two dental messages deliberately disagreeing about the appointment date. These are fictional evaluation inputs, not completed household cases.
Codex helped implement the authenticated dashboard, persistent task schema, multimodal analysis and failure behavior. Runtime GPT-5.6 produces structured task data, but text evidence is checked against the submitted source before a live response is accepted. Conflicting sources remain separate so the interface can expose disagreement instead of choosing a convenient date.
The build stores derived tasks and evidence, not original uploaded documents. Live and deterministic-fallback provenance are persisted and labeled. A completion update is scoped to the authenticated user and survives refresh. Generated messages remain drafts; calendar actions produce downloaded files rather than silently contacting an external service.
The creator reports verification of responsive behavior, failure states and build output. Known limits remain material: account deletion and some recovery flows are incomplete, re-analysis creates a new run instead of deduplicating tasks, and raw documents cannot be reopened. The useful result is an evidence-aware prototype with visible human checkpoints, not autonomous payment, form submission or professional advice.
Prompt
Suggested prompt (written by Agents In Use, not the author's)
Source: Devpost · Nithish K
Frequently asked questions
Were the sample household cases real?
No. The source explicitly labels all bundled people, accounts and events fictional.
Are uploaded originals archived?
No. Only derived task information and excerpts persist, so raw documents cannot be reopened later.
Does it send messages or pay bills automatically?
No. Messages stay drafts and calendar outputs are downloaded files.
More details
About this account
Nithish K defined a source-grounded, approval-first prototype and supplied fictional conflict samples. Codex implemented validated analysis, user-scoped persistence and separately labeled live and fallback modes.
Firsthand dated prototype account opened. Codex build and GPT-5.6 runtime roles separated. Sample inputs are explicitly fictional; incomplete deletion, recovery and deduplication flows are disclosed, not hidden behind feature claims. Suggested prompt written by Agents In Use; no original task prompt was found in the verified text.
Credit: Nithish K — Devpost
Source published
Added to Agents In Use
Source checked
Tools and topics
Codex · OpenAI Responses API · Cloudflare D1 · Better Auth · JSON Schema
document conflicts · evidence excerpts · approval-first prototype
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.