How to use CODEX to personalise and index jobs from seek.com.au - By Ian Han

prompt as follow: Use $skill-creator to create and install a reusable Codex skill named

"job-index-[year]".

Purpose:

Create a daily job index from SEEK for jobs advertised in [LOCATION].

The skill must collect jobs posted today or within the last 24 hours,

save them in CSV and Excel formats, and highlight jobs that suit the

user.

Before creating the skill, ask the user for any missing information:

- Their name

- Location and timezone

- SEEK search URL or desired location

- Workspace folder

- Resume/profile/work-experience file locations

- Target job fields

- Preferred employers or organisation types

- Roles or occupations that should normally be excluded

- Desired schedule, if automation is requested

Create the skill under the user's Codex skills directory using a

portable name such as:

job-index-[year]

The skill should contain:

1. SKILL.md

2. agents/openai.yaml

3. references/matching-rules.md

4. scripts/seek_daily_job_scan.py

The Python script must:

- Use SEEK's public-facing search results.

- Apply SEEK's Today or last-24-hours filter.

- Fetch every results page reported by SEEK.

- Extract structured job information from the public page data.

- Deduplicate jobs by SEEK job ID.

- Sort jobs from newest to oldest.

- Use the user's local timezone correctly, including daylight-saving

changes.

- Read the user's resume, experience, skills, eligibility constraints,

and target-role files when available.

- Classify jobs as "Strong fit", "Possible fit", or not highlighted.

- Explain every highlighted result in a "Match Reason" column.

- Produce a date-stamped CSV containing the complete job index.

- Produce a date-stamped XLSX workbook.

- Highlight Strong fit and Possible fit rows in yellow.

- Include job ID, posting time, local posting time, title, company,

location, work type, work arrangement, salary, and SEEK URL.

- Compare the number of unique extracted jobs with SEEK's reported

result count.

- Never claim the result is complete when those counts do not match.

- Retry temporary network failures and report parsing failures clearly.

- Avoid modifying the user's resume or profile documents.

- Use configuration values rather than hardcoded names, dates,

usernames, timezone offsets, or absolute paths.

Matching must consider three sources:

1. The user's resume, experience, skills, eligibility, and target roles.

2. The user's nominated job fields:

[INSERT TARGET FIELDS]

3. The user's preferred employers or organisation types:

[INSERT PRIORITY ORGANISATIONS]

Put detailed Strong fit, Possible fit, exclusion, and priority-employer

rules in references/matching-rules.md. State clearly that yellow means

"review this job", not "automatically apply".

Write a concise SKILL.md with a comprehensive trigger description.

The skill should trigger when the user asks to scan SEEK, find jobs

posted today, create a job spreadsheet, highlight suitable roles,

match jobs against their profile, or manage the related automation.

Create agents/openai.yaml with:

- display_name: "Job Index [Year]"

- short_description: "Index and highlight suitable SEEK jobs"

- default_prompt: "Use $job-index-[year] to scan today's SEEK jobs in

[LOCATION] and create a highlighted Excel index."

Test the Python script with a live public SEEK search. Validate the

finished skill using the skill-creator validation tool. Install it in

the user's Codex skills directory and report its installed location.

Do not create a scheduled automation unless the user separately

authorises it. If requested, use Codex's supported automation system

and schedule it in the user's local timezone.