Get Bullhorn API access via OAuth 2.0, then use it to audit and clean up Herefish workflows, remove unused ones, and add missing automations. One task, two phases: credentials first, cleanup second.

OAuth: Every Bullhorn automation downstream (job sync, candidate mapping, event bus) needs programmatic API access. Currently tokens are obtained manually — this blocks everything.

Herefish: A new workflow gets created for every mailout/text campaign, leaving a pile of abandoned workflows. Hard to tell what's active vs unused. Cleaning this up means Herefish handles routine stuff (rejection emails, KSC acknowledgements) automatically instead of manually.

Yes — the gateway to everything Bullhorn. OAuth enables all API automation. Herefish cleanup is a quick win since it's already built into Bullhorn — no cost or integration work, just tidy it up. Both reduce admin time directly.

Phase 1 — OAuth: Cloud Function (bullhornOAuth) deployed and ready (callback, token storage, auto-refresh). ✅ API credentials received from Bullhorn on 17 Jun 2026.

Credentials (saved to .env):
• Client ID: 6e04b191-410d-43e1-a78a-13b09ce25575
• Client Secret: 2tuujyViVxOc3F83YLSRZlvW
• API Username: limeauto.restapi
• Data center: 60 (CLS60, APAC/Singapore)
• Refresh tokens: Enabled

⚠️ Still needed: Generate API password in Bullhorn admin (Admin → Users → limeauto.restapi → generate password). Requires Paul's admin access. This is the one blocker before we can test the auth flow.

Bullhorn Support Tickets:
• Ticket 06070357 (Mosia) — auto-closed, never responded.
• Ticket 06097317 (15 Jun 2026) — ✅ API access granted. Gopi Modi sent credentials 16 Jun 2026.
• Ticket 06071354 (Dan) — tearsheet ticket, can be closed.

Phase 2 — Herefish Audit (completed 2026-06-09):
• 969 active automations — massive bloat, most are stale
• 50 hibernated automations
• Jobs from Jan/Feb 2026 still "Active" and emailing candidates (closed 4-5 months ago)
• No groups/organization — almost all have empty Groups column
• Duplicates found: e.g. "PM QCS 7/4/26 capital works" (57 active) + "pm qcs 7/4/26" (61 active)
• Broken automations: "Happy Birthday!" (0/0 never fired), "AR - Auto response" (0 active, 754 total — may not trigger)
• No cleanup process — automations stay Active forever
• Types: mostly Candidate-based, 1 Submission-based, 1 Placement-based (27-step onboarding)
• Most have 2 steps (initial + follow-up), some 3-4 steps

Phase 1 — OAuth:
1. Paul: Send follow-up email to Mosia (ticket 06070357) — ✅ Done (15 Jun 2026). New ticket submitted.
2. Jackson: Reply to Bullhorn support (ticket 06097317) with OAuth config — ✅ Done (16 Jun 2026).
3. Bullhorn: provisions dedicated API user + sends Client ID & Client Secret — ✅ Done (16 Jun 2026). Credentials received.
4. Jackson + Paul: Log into Bullhorn admin → Admin → Users → limeauto.restapi → generate password. Add BULLHORN_API_PASSWORD to .env. (BLOCKER — needs Paul's admin access)
5. Test the full OAuth flow end-to-end
6. Build Bullhorn data sync (jobs, candidates, placements → Firebase)

Phase 2 — Herefish Cleanup (IMMEDIATE):
1. Hibernate all automations for jobs closed 30+ days ago — est. ~600-700 of 969 are stale
2. Find and remove duplicates (search for similar names/dates)
3. Fix "Happy Birthday" automation (0/0 = broken trigger)
4. Check "AR - Auto response" trigger configuration

Phase 2 — Herefish Organization (SHORT TERM):
5. Create Groups: by client (TMR, QCS, Seqwater, etc.) and by status (Active Jobs, Closed, Evergreen)
6. Standardize naming: [Status] [Role] [Client] [DD/MM/YY]
7. Create automation templates for common role types

Phase 2 — Herefish Ongoing:
8. Set up weekly review process to pause/hibernate old automations
9. Set up alerts when automations have 0 active for 30+ days
10. Build dashboard view of Herefish automation health metrics

📋 Copy

Clean and enrich the Bullhorn database — remove stale/inaccurate candidate records, update contact details, fill in missing fields — so that sourcing and ranking algorithms work against accurate data.

Paul said it directly: "We really need to clean data." Bullhorn has old resumes, incorrect contact details, and outdated skills. If we run AI sourcing against dirty data, we'll get garbage results. This is the foundation that makes candidate ranking trustworthy. Also: the team needs to commit to "everything goes in Bullhorn" — meeting notes, candidate interactions, all of it.

Yes — prerequisite for good sourcing. We could skip it and accept lower quality results, but Cass already doesn't trust the automation. Running it against bad data will confirm her fears. Clean data first, then source.

Not started. Blocked on: Bullhorn API access. Once OAuth is live, we can programmatically audit records.

1. Wait for Bullhorn API credentials
2. Export candidate records via API — identify duplicates, stale records, missing fields
3. Use AI to enrich records (LinkedIn profile matching, contact verification)
4. Remove or archive candidates not updated in 12+ months
5. Establish "everything in Bullhorn" team convention
6. 🔔 Note: Hermes can use web search + LinkedIn to enrich profiles automatically once API access is live

Turn Seek's ad optimisation best practices into an automated system that creates high-performing job ads, monitors their performance, and adjusts them — so consultants get better candidates without manually tweaking ads. Also review Lime's internal application emails against Seek's recommendations.

From the Seek meeting (17 Jun 2026):

Market context: QLD job ads down 10% but applications per ad are UP — more competitive. Less ads = each ad needs to stand out more. Seek's AI reads every ad and matches candidates based on content quality.

How Seek's AI works:
• Seek AI reads ad content, job title, salary, skills, requirements on posting
• Analyses candidate behaviour: scroll pauses, hovers, shortlists, successful placements
• Positions different ads to different candidates based on their behaviour profile
• Gives on-platform and off-platform recommendations to candidates
• Provides explainability for why a candidate is a strong fit

Ad performance funnel: Exposure (title) → Impressions (title+salary+summary) → Clicks (bullet points) → Completed applications

Key optimisation levers:
• Job title: Add specialty after dash (e.g., "Project Manager — Finance Transformation"). Specific titles get better exposure than generic ones. Put AO banding at END of title if required.
• Salary field (80 chars): Always put something meaningful — hybrid working, competitive rate, location perks. 7 ads had blank salary fields.
• Short summary (150 chars): Use for benefits — CBD location, hybrid, career progression, free parking
• Benefit bullet points: Need branding package (basic+branding ~$80 extra). Shows work-life balance, flexibility, salary. Currently all ads are basic ads with no benefit bullets.
• Ad body: 2-3 sentences about role, 5 bullet duties, 5 bullet requirements. Short, scannable, mobile-first. Big walls of text = candidates skip.
• AI writing tool: Seek has built-in AI to help write ads. Can upload PD and it generates content. Wendy could use this.

Lime's current ad performance: Great exposure and impressions (better than benchmarks). But some ads have very low click-to-impression conversion — wrong candidates seeing the ad. Some ads have red completion rates (30% less than benchmark). 7 ads with blank salary fields.

Yes — high impact, relatively low effort. Better ads = better candidates = less time sorting through unsuitable applicants. This is a multiplier on everything else we're building. The 72-hour analysis cron is especially valuable — currently nobody reviews ad performance after posting.

Not started. Seek meeting completed 17 Jun 2026. Documentation research needed — Seek has a developer API (developer.seek.com) with GraphQL, Job Posting, Ad Performance, and Optimised Apply endpoints. Also need to review Seek's hiring advice articles at au.employer.seek.com/hiring-advice.

Seek Developer API capabilities found:
• Job Posting API (GraphQL) — create/update/close job ads programmatically
• Ad Performance API — track ad metrics
• Ad Sync — sync positions between ATS and Seek
• Enhanced Job Posting — drafts, position input, display panels
• Partner tokens & browser tokens for auth

Immediate blocker: Seek login session expires on the Mac. Need persistent auth before any cron can work. May be solvable via Seek's Partner API instead of browser automation.

Part 1 — Seek Ad Creation Agent (Cron):
1. Deep research: read ALL Seek documentation — developer.seek.com API docs, au.employer.seek.com/hiring-advice articles, Laws of Attraction data
2. Create a Hermes skill with Seek's ad writing guidelines as system prompt
3. Build cron agent: when new job arrives (from FG webhook or manual trigger), auto-generate optimised Seek ad content using Seek's best practices
4. Auto-fill: title with specialty, salary field with meaningful content, short summary with benefits, structured body (2-3 sentences + 5+5 bullets)
5. Draft for consultant review before posting (don't auto-post initially)

Part 2 — 72-Hour Ad Analysis Cron:
1. After ad is posted, schedule a cron job to run 72 hours later
2. Pull ad performance data: impressions, clicks, applications, completion rate
3. Compare against Seek benchmarks and similar ads
4. Identify issues: low impression rate (title/salary problem), low click rate (bullet points/summary), low completion rate (body too long/vague)
5. Auto-suggest specific changes (e.g., "Add specialty to title", "Salary field is blank — add hybrid working perk")
6. If authorised, auto-update the ad via Seek API
7. Track improvements over time — did the changes improve performance?

Part 3 — Application Email Review:
1. Review Lime's current application emails (the ones candidates receive when they apply)
2. Compare against Seek's documentation on candidate experience
3. Check: are we providing explainability? Are we nudging candidates to similar roles?
4. Recommend changes to align with Seek's platform behaviour

Part 4 — Seek Documentation Skill:
1. Compile all Seek documentation into a Hermes skill
2. Include: ad writing guidelines, Laws of Attraction data, ad performance funnel metrics, AI matching rules, salary/title best practices
3. This skill becomes the knowledge base for both the ad creation agent and the analysis cron
4. Update skill whenever Seek publishes new guidance

Developer API:
• developer.seek.com — GraphQL API, Job Posting, Ad Performance, Ad Sync, Optimised Apply
• Ad Performance API — track ad metrics programmatically
• Job Posting API — create/update/close ads via GraphQL

Hiring Advice:
• au.employer.seek.com/hiring-advice — ad writing tips, market insights
• "4 tips to attract the right candidates" (14 May 2026)
• "Are candidates seeing your job ad? Make these changes" (16 Apr 2026)
• "Job titles: why they matter more than you think" (18 Feb 2026)
• "Laws of Attraction" — data on what candidates want (work-life balance, salary, flexibility, career progression)

From Seek Rep (Kayla) at meeting:
• Seek has the "biggest responsible AI team in the southern hemisphere"
• Candidate behaviour tracking: scroll pauses, hovers, shortlists, successful placements
• Seek AI writing tool built into the platform
• Auto-apply detection work in progress (spam/bot filtering)
• Partner Management team works with ATS integrations (60+ partners)
• Contact: Will Hoyt (partner management) for API integration questions