Works with the AI Agents you already use
OpenAI Prompt Objects & Evals Shutdown Fix (Nov 30, 2026): Move pmpt_ Prompts to Code, Evals to Promptfoo
Any app that calls the Responses API with `prompt: { id: "pmpt_..." }`, or any release process gated on OpenAI Evals, breaks unless it moves first.
$1
OpenAI Prompt Objects & Evals Shutdown Fix (Nov 30, 2026): Move pmpt_ Prompts to Code, Evals to Promptfoo
Example session with this skill installed
Our support bot calls client.responses.create with prompt={"id": "pmpt_68ab12cd34ef", "version": "4"} and our onboarding route reads a prompt ID from ONBOARDING_PROMPT_ID. A release script runs client.evals.runs.create. What breaks on November 30 and can you migrate it?
- Read your context and instructions
- Compiled the openai prompt objects
OpenAI Shutdown Migration Report — support-app — 2026-11-10
Verdict
NOT READY → READY after changes below: 5 blockers fixed, 3 model warnings scheduled.
Inventory (from scanner, exit code 1 before / 0 after)
| ID | Kind | Location | Deadline | Status |
|---|---|---|---|---|
| P1 | prompt-call + prompt-id | app/support.py:6 (pmpt_68ab12cd34ef, v4) | 2026-11-30 | migrated → prompts/support_reply.json |
| P2 | prompt-call | app/route.ts:2 (ID from env) | 2026-11-30 | migrated → prompts/onboarding.json |
| P3 | prompt-id-config | .env.example:1 | 2026-11-30 | removed |
| E1 | eval-api | scripts/run_eval.py:1 | 2026-11-30 | rebuilt → promptfooconfig.yaml |
| M1 | model-id gpt-image-1.5 | app/route.ts:4 | 2026-12-01 | open: replace with gpt-image-2.6-flare or -sunburst |
| M2 | model-id gpt-5-mini | app/support.py:12 | 2026-12-11 | open: official replacement gpt-5.6-terra |
| M3 | model-id whisper-1 | app/support.py:15 | 2027-02-26 | open: gpt-transcribe or gpt-live-transcribe |
Verification
| # | Check | Result |
|---|---|---|
| V1 | No pmpt_/v1/prompts left | PASS (re-scan: 0 prompt findings) |
| V2 | Rendered prompt = dashboard v4 | PASS (diff empty) |
| V4 | Loader rejects missing/unused vars | PASS ("missing variables ['issue']", "unused variables ['extra']") |
| V6 | promptfoo echo dry run | PASS (vars rendered, no {{ left) |
| V7 | CI gate red on failing case | PASS (exit 100 with failing case, 0 without) |
Open items
M1–M3: run the promptfoo suite against each replacement model before switching production.
Connects securely to your tools. The creator never sees your data.
About this skill
The deadline
On November 30, 2026 OpenAI shuts down the v1/prompts API and reusable prompt objects, the Evals dashboard and API (read-only since October 31), and Agent Builder (official deprecations page, checked 2026-11-10). Any app that calls the Responses API with prompt: { id: "pmpt_..." }, or any release process gated on OpenAI Evals, breaks unless it moves first.
What this skill does
- Scans the repo with a bundled, read-only, standard-library Python scanner:
pmpt_IDs in code and.envfiles, Responses API calls with a prompt object,v1/promptsandclient.evalscalls, possible Agent Builder workflow IDs, and every OpenAI model ID with a shutdown date from Dec 2026 to Feb 2027. Exit code 1 when blockers exist, so it can gate CI. - Inventories each dependency with file:line, pinned version, variables and deadline.
- Recovers each prompt's exact published content plus the settings hidden in the prompt object (model, tools, text-format schema, reasoning, sampling) that your code never showed.
- Moves prompts into code as one JSON file per prompt, read by strict loaders for Python, TypeScript and PHP/Laravel that fail on missing, unused or unreplaced variables.
- Rebuilds evals in promptfoo using the same prompt file as the app, with a grader-by-grader mapping table (string check, text similarity, label model, score model, Python) and a free
echodry run that checks rendering without API cost. - Verifies with an 8-point PASS/FAIL check (no IDs left, rendered prompt equals the dashboard version, settings carried over, CI gate proven red on a failing case) and writes
OPENAI-SHUTDOWN-REPORT.md.
Why not just prompt it yourself
General models do not know these dates, the official replacements, or that a prompt object can silently carry tools and output schemas. The loaders and promptfoo config in this skill were run and tested on a sample repo on 2026-11-10 (promptfoo validate passed; eval exit codes observed: 100 on failure, 0 on pass).
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for LLM Ops.
Cannot read your OpenAI dashboard: you (or the agent with your help) must copy each prompt's published content and grader definitions before November 30. Agent Builder workflows are detected and listed, not rebuilt. The grader-to-assertion mapping is a recommended equivalent, not an official table, so LLM-judged graders must be re-validated. Prompt IDs resolved at runtime from a database or no-code tool are not found by the scanner. Dates and replacements were checked on 2026-11-10 and can change; re-check the deprecations page after 2026-12-10.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 7 days ago
- Accounts
- OpenAI API key for real eval runs
- Software
Python 3.8+ (scanner)Node.js for npx promptfoopromptfoo
Node.js
Permissions
Allowed Hosts
File Scopes
The bundled scanner is read-only, standard-library only and makes no network calls.