Job Autopilot — Search Agent

Searches LinkedIn (and optionally company career pages) for roles matching your criteria, applies hard filters, and writes results into a structured tracker file.

Setup

All env vars below are set by setup.sh (from jobautopilot-bundle). If you install standalone, add to ~/.openclaw/users/<you>/config.sh:

CODEBLOCK0

Initialize the tracker if it does not exist yet:

CODEBLOCK1

Read at session start

Every session, before searching, the agent must read in order:

1. 1. $RESUME_DIR — build a candidate profile from the user's resume pool (see below)
2. INLINECODE4 — check existing entries to avoid duplicates
3. INLINECODE5 — pick up context from previous sessions

How to read the resume pool

Scan all files in $RESUME_DIR (.docx, .pdf, .md, .txt). From them, extract and record:

• - Skills — programming languages, tools, frameworks, domain knowledge, certifications
• Titles held — past and current job titles, seniority level
• Industries / asset classes — sectors the user has worked in
• Preferred roles — infer from the most recent or most polished resume; note any target roles the user has written explicitly
• Location / remote preference — extract from contact header or any explicit statement
• Seniority signals — years of experience, scope of responsibility, team size managed

Synthesize these into a short candidate profile (keep it in working memory for this session). Use the profile to:

• - Derive search keywords (e.g. titles, skills, domain terms)
• Set the seniority filter (reject roles that are clearly too junior or too senior)
• Prioritize industries and company types that match past experience
• Skip roles where the user clearly lacks the stated hard requirements

Search behavior

Use the browser tool with profile search. Search LinkedIn Jobs as the primary source. Also search company career pages for target employers when useful.

Keyword combinations to try (mix and match from config):

CODEBLOCK2

Repeat with multiple keyword pairs. Log each search query in $JOB_SEARCH_HANDOFF so future sessions do not repeat the same queries.

Hard filters — reject if ANY applies

• - Location is not $JOB_SEARCH_LOCATION (or explicitly remote)
• Posted more than $JOB_SEARCH_MAX_AGE_DAYS days ago
• Salary listed and below INLINECODE15
• Role is clearly junior (< 3 years required) unless explicitly configured otherwise
• Duplicate of an existing tracker entry (same company + title)

Tracker format

Append one row per kept result to $JOB_SEARCH_TRACKER. Use this format exactly:

CODEBLOCK3

Status values: shortlist | rejected | INLINECODE19

Both kept and rejected results must be recorded. Every kept result must include the exact job URL.

Handoff notes

After each search session, update $JOB_SEARCH_HANDOFF with:

• - Queries run this session
• Date range of listings found
• Any platforms or companies worth revisiting
• Anything unusual encountered

Optional: Find hiring managers (user-initiated only)

This step is only performed when the user explicitly asks (e.g. "find the hiring manager", "who should I contact at X"). It is never run automatically.

When requested, use the browser tool to look up the relevant person via:

• - LinkedIn — search for recruiters or engineering managers at the company
• Company website — check the careers page or team page for contact information

Record any found contacts in the tracker's Notes column.

Scope

This skill covers search, screening, and optional hiring manager outreach. Do not tailor resumes, write cover letters, or submit applications. Hand off shortlist entries to the jobautopilot-tailor skill.

Support

If Job Autopilot saved you time: paypal.me/ZLiu308