This skill monitors public X posts from Codex lead thsottiaux, detects possible Codex reset signals, and outputs findings through Codex Automation. It helps users learn about reset-related changes before a reset happens, so they can spend the remaining quota in their subscription more intentionally, including switching to fast mode when useful.
This skill does not require you to register any API and does not create extra costs, though running Codex Automation still consumes your Codex quota.
How To Use?
Step 1: Create A Codex Project First
Create a dedicated Codex Project for this monitor, for example Codex Reset Watchdog.
Step 2: In That Project, Copy The Prompt To Install The Skill And Create The Automation
Open the Project you just created, start a new chat inside that Project, then paste the prompt below. Codex will install this skill by itself.
⚠️ Note: do not run the prompt below in a normal Chat. You must run it inside the Project you just created.
Quietly install, initialize, and enable codex-reset-watchdog:
https://github.com/thinkingjimmy/codex-reset-watchdog
Only message me mid-run if you need my approval or hit a blocker that I must resolve.
Otherwise, do not output progress narration, tool parameter details, command attempts, retry details, raw JSON, or state file contents. Complete the tasks below yourself, then give only a concise setup summary.
Tasks:
1. Prefer Codex's skill installation workflow for this GitHub repo, using the skill name codex-reset-watchdog. If no installer is available, clone the repo.
2. Find the installed or cloned source directory and confirm it contains SKILL.md, scripts/check_once.mjs, references/automation-prompt.md, and .codex/config.toml.
3. Prepare the runtime files in the current workspace root: SKILL.md, README.md, README.zh-CN.md, env.example, .codex/, agents/, references/, scripts/, and images/. Preserve the existing .git directory, do not create a nested repo, and do not overwrite local env or .env.
4. In the current workspace root, confirm SKILL.md, scripts/check_once.mjs, references/automation-prompt.md, and .codex/config.toml exist.
5. In the current workspace root, run node scripts/self_test.mjs.
6. Run node scripts/check_once.mjs --prime-state --json to create the baseline state.
7. Run node scripts/check_once.mjs --dry-run --json to confirm the Dayclaw public source, JSON parsing, and state dedupe.
8. If node scripts/check_once.mjs fails because of sandbox/network permissions, request permission only for the narrow node scripts/check_once.mjs entrypoint and rerun; do not request full access. If DNS/HTTPS or state writes still fail, summarize them as operational issues, not as reset/no-reset conclusions.
9. Read the full contents of references/automation-prompt.md as the Automation prompt.
10. Use `automation_update` to create or update a cron/project Automation named Codex Reset Watchdog. If the Automation tool is not already available, search for `automation_update` first. Look for an existing Automation with the same name or id and update it instead of creating a duplicate. For create/update, use these fields exactly, and do not replace field names or values with UI labels: `mode=create` or `mode=update`, `kind=cron`, `name=Codex Reset Watchdog`, `rrule=FREQ=HOURLY;INTERVAL=1`, `status=ACTIVE`, `executionEnvironment=local`, `cwds=[current workspace root]`, `prompt=[full contents of references/automation-prompt.md]`, and `reasoningEffort=medium`. Do not pass `command`, `permissions`, `cwd`, `frequency`, `schedule`, `destination`, or thread/heartbeat fields. If the tool requires `model`, use the current Codex default model. If this field set is still rejected, summarize it as a setup blocker; do not repeatedly trial-create invalid Automations.
11. Final summary only: runtime directory, install source directory, self-test, prime/dry-run status, state.path, source health, Automation name/status/cadence/execution/cwd/prompt source, and Run Now expectation. Do not paste raw JSON.
Step 3: Test The Automation
After creation, click Run Now (shown as 2 in the image) on the Automation (shown as 1 in the image) detail page. As expected, you should see the latest chat output inside the Codex Reset Watchdog Project.
When a possible reset signal appears, the result may look like this (example):
🚨 Actionable Codex reset ahead: paid ChatGPT Codex limits are scheduled to reset. Reset timing: 2026-06-03 morning (Asia/Shanghai).
| Time | Evidence | Reset timing | Actionability | Link |
|---|---|---|---|---|
| 2026-06-02 22:15 Asia/Shanghai | Said limits are "resetting tomorrow morning". | 2026-06-03 morning | 🚨 future | https://x.com/example/status/3 |
Want To Monitor More Sources?
If you want to monitor accounts beyond thsottiaux, or other social platforms such as Reddit, Hacker News, or Product Hunt, you can try Dayclaw.
FAQ
Q: Why not use a normal chat?
Because Codex Automation sandbox restrictions can prevent required install, check, and state-write operations from completing in a normal chat. That is why this flow uses a Project.
Q: How do I set the timezone?
With the default configuration, Codex monitors @thsottiaux, writes state to var/state.json, and displays times using the Automation runtime environment. If you want findings to use another timezone, create an env file and override the default REPORT_TIMEZONE:
- Duplicate
env.example. - Rename the copy to
env. - Change only the fields you actually need. To force a report timezone, set
REPORT_TIMEZONEto an IANA timezone:
REPORT_TIMEZONE=America/Los_Angeles
Common examples: Asia/Shanghai, America/Los_Angeles, America/New_York, Europe/London, Europe/Berlin, UTC.
Leave REPORT_TIMEZONE blank to use the Automation runtime/user timezone. This is the recommended default.
Skill Layout
This repo is a single skill directory: SKILL.md sits at the root, with optional agents/, references/, and scripts/ beside it. To embed this skill inside another repo, copy this directory to .agents/skills/codex-reset-watchdog/.
codex-reset-watchdog/
.codex/
config.toml # Minimal Codex permission profile
rules/
codex-reset-watchdog.rules # Command-level network fallback for older sandbox mode
SKILL.md # Skill metadata and operating instructions
README.md # English guide
README.zh-CN.md # Chinese guide
env.example # Visible configuration template
.gitignore # Ignores caches and local state
agents/
openai.yaml # Optional Codex skill display metadata
images/
banner.png # README banner image
previous-runs.png # Automation run screenshot
references/
automation-prompt.md # Prompt used when creating Codex Automation
deployment.md # Operations checklist
llm-judge-rubric.md # Rules for LLM review of review_items
scripts/
check_once.mjs # Zero-dependency Automation entrypoint
self_test.mjs # Local deterministic self-test