Plan C: PPMO+ plus GHL plus OpenAI Codex (Cody) Autonomously Controlling Linked-Helper

Authors: ExCom+ Unified (Betty chair, Willemena, Sade, Gemma, Vesta) with Clarise and Claudia review Marc-directive: 2026-05-18 PM, "How do we extend all thing EGP PPMO+ into EGP GHL + OpenAI Codex fully controlling Linked-Helper" Status: Draft v1, awaiting Marc ratify Date: 2026-05-18 Jira candidate epic: SWAT-PPMO-CODEX-LH-01

1. Executive Summary

Linked-Helper is the canonical LinkedIn automation engine for PPMO-LH (50/day saved-jobs cadence; 110 recruiter touches; Sales Nav weekly sweep). Marc wants Cody (PPMO-OA, OpenAI Codex) to take directive input and execute Linked-Helper campaigns autonomously, with GHL as the data plane (lead lists in, conversation captures back). The recommendation is Path D (Hybrid): keep Linked-Helper as the LinkedIn-action engine for the workflows it already handles well (connection requests, drip messages, profile visits, Sales Nav exports) and add custom Playwright for residual cases Linked-Helper does not cover (saved-jobs sweep capped at 50/day, job-tracker dup handling, custom InMail variants). Cody orchestrates both via a unified directive interface and writes lead and conversation data to GHL. Estimated runbook 30 working days, 4 sprints, with first autonomous campaign live by day 9.

2. Current State

[Marc directive in chat]
        |
        v
[Manual: Marc opens Chrome, launches Linked-Helper]
        |
        v
[Linked-Helper extension runs] --- CSV in/out --- [tools/linkedhelper_*.py]
        |
        v
[Google Sheets job tracker (1Bk_…)] --- [tools/linkedhelper_ghl_bridge.py] --- [GHL]

Linked-Helper is a Chrome extension; controlled via UI and a local REST surface (port 9080 by default when configured).
Existing tools: linkedhelper_automation.py, linkedhelper_ghl_bridge.py, linkedhelper_gsheets_sync.py, supabase_to_linkedhelper.py, egp_a4_linkedhelper_500_build.py.
50/day LinkedIn cap, 50-100/day Dice cap, hourly saved-jobs cron, weekly Sales Nav cron (per feedback_persistent_autonomous_off_hours_operation_canonical_2026_05_18.md).
Marc still launches campaigns manually; Cody not yet wired.

3. Target State

[Marc directive (chat or Directive Capture widget)]
        |
        v
[Cody Intake API (shared with Plan B)]
        |
        v
[Linked-Helper Action Router]
        |              \
        v               v
[Linked-Helper local REST API]    [Playwright on Linked-Helper UI + LinkedIn]
              \                  /
               \                /
                v              v
              [Lead + conversation capture]
                       |
                       v
              [GHL contacts + opportunities + Conversations]
                       |
                       v
              [Memory + Confluence + ship report to Marc]

4. Runbook

Task	Owner	Effort (days)	Dependencies	Definition of Done
T-01 Audit current Linked-Helper config, workflows, queues, daily caps	Willemena	1	none	Audit doc in `04-operations`
T-02 Enable and validate Linked-Helper local REST API (port 9080)	Vesta	1	T-01	`/api/workflows` returns 200
T-03 Wrap Linked-Helper REST in Python client (`egp_linkedhelper_api_client.py`)	Sade	2	T-02	12 methods callable
T-04 Playwright executor for non-API Linked-Helper actions (start/stop campaign, edit message template)	Sade	3	T-02	6 actions automated
T-05 Cody Intake integration (shared with Plan B Cody service)	Cody + Claudia	1	Plan B T-01	Linked-Helper directives routed to LH router
T-06 Directive intent labels for Linked-Helper (connect, message, search, export, audit)	Claudia	2	T-05	50-directive labelled set, 95 percent accuracy
T-07 Lead-list provisioning: GHL smart-list -> Linked-Helper queue	Willemena	2	T-03	110-recruiter list provisioned end to end
T-08 Conversation capture: Linked-Helper replies -> GHL Conversations tab	Willemena	2	T-03	Reply visible in CRM within 60s
T-09 Saved-jobs custom Playwright bot (Linked-Helper does not cover this well)	Sade	3	T-04	50/day cap honored, dup handler fires
T-10 Daily-cap enforcement: persistent counters at `.tmp/linkedhelper_daily_counter.json`	Clarise	1	T-03	Counter persists across restarts
T-11 Pre-action guardrails (no message to existing customers, no duplicate connect within 90 days, em-dash lint on message text)	Betty	2	T-03	HARD-FAIL on violation
T-12 Audit log + ship report (campaign-level + per-action)	Clarise	1	T-03	JSONL + daily email to Marc
T-13 Off-hours scheduling integration (existing launchd plists)	Vesta	1	T-10	LH campaigns continue when Marc OOO
T-14 Telemetry dashboard pane (campaigns active, leads contacted, response rate)	Gemma	2	T-12, Plan A	Pane live in Command Center
T-15 Approval gate for new message templates and new campaign creation	Betty + Sasha (Skeptic)	1	T-11	Approval flow tested
T-16 Acceptance test (Marc directive -> live campaign in under 10 min)	Betty	1	all	3 directives shipped end to end
T-17 Hardening: Chrome restart, extension reload, LinkedIn challenge handler	Sade	3	T-16	Chaos test passes
T-18 Cutover and runbook	Willemena	1	T-17	Confluence runbook live
Slack	n/a	2	n/a	reserved

Total: 30 working days, 4 two-week sprints.

5. SWOT

Strengths	Weaknesses
Linked-Helper already running, known workflows, persistent off-hours	Linked-Helper local API requires Chrome running and extension active
50/day cadence + dup handling already canonical	Some LH actions only exposed via UI (need Playwright)
GHL bridge already exists	LinkedIn challenge or 2FA can pause campaigns
Daily counter pattern already battle-tested	Cody not yet a long-running service (depends on Plan B T-01)

Opportunities	Threats
Removes Marc from manual campaign launch	LinkedIn ToS and account restrict risk
Unlocks 5x to 10x saved-jobs volume	Linked-Helper vendor sunset risk
Pattern reusable for Sales Nav + Recruiter	Chrome auto-update breaking extension

6. Five Solution Paths

Path A: Cody controls Linked-Helper via its native REST API

Approach: only call documented Linked-Helper REST endpoints (port 9080).
Pros: fastest, cleanest, fully scriptable.
Cons: limited surface (campaign start/stop, queue read, basic stats); does not cover template editing, saved-jobs sweep, custom InMail variants.
Effort: 12 days.
Recommended: as the 60-percent path.

Path B: Cody invokes Linked-Helper via Playwright (controlling the extension UI)

Approach: drive the Linked-Helper Chrome extension and LinkedIn pages via Playwright.
Pros: covers every action a human can do.
Cons: very fragile, Chrome must stay running, breaks on extension or LinkedIn UI updates.
Effort: 22 days.
Recommended: only for actions API cannot do.

Path C: Replace Linked-Helper with a custom Playwright bot

Approach: deprecate Linked-Helper; build EGP-native LinkedIn bot end to end.
Pros: full control, no vendor risk, no extension surface.
Cons: throws away years of Linked-Helper hardening (rate limits, throttles, LinkedIn-safe patterns); large attack surface for LinkedIn detection.
Effort: 60+ days.
Recommended: no, throws away working asset.

Path D: Hybrid Linked-Helper for known workflows plus custom Playwright for residual (RECOMMENDED)

Approach: keep Linked-Helper as the engine for connect, drip, profile visit, Sales Nav export; add custom Playwright for saved-jobs sweep, custom InMail, and any future workflow LH does not cover.
Pros: preserves Linked-Helper hardening, removes the gaps, hedges vendor risk.
Cons: two execution surfaces to maintain.
Effort: 30 days (full runbook).
Recommended: YES.

Path E: Decompose to GHL workflows where possible, Linked-Helper for residual

Approach: push as many lead and message workflows as possible into GHL native automations; Linked-Helper handles only LinkedIn-specific actions.
Pros: leverages GHL retry, scheduling, conversation features.
Cons: GHL does not natively touch LinkedIn; only the data plane benefits.
Effort: 18 days.
Recommended: as a complement to Path D, not a substitute.

7. Top-3 Risks (Path D)

LinkedIn account restriction. Mitigation: respect 50/day cap (already canonical), warm cadence on new actions, 2FA challenge handler at T-17, daily reputation check.
Linked-Helper extension breaks. Mitigation: pinned Chrome version, pinned extension version, weekly synthetic test, fallback to Playwright bot for the affected workflow.
Customer in a connect list. Mitigation: T-11 guardrail cross-references GHL customer table and excludes anyone marked is_customer=true or having open opportunities.

8. Jira Recommendation

Project: SWAT
Epic: SWAT-PPMO-CODEX-LH-01 "Cody autonomous control of Linked-Helper"
First sprint stories:
T-01 Audit current LH config
T-02 Enable LH local REST API
T-03 Python client wrapper
T-07 Lead-list provisioning end to end
T-09 Saved-jobs custom Playwright bot
Confluence parent: EGPSWAT space, "Cody Linked-Helper Executor".

9. Where We Left Off

01-strategy/strategy/2026-04-28-EGP-Linked-Helper-Status-Relaunch/ (relaunch spec)
tools/linkedhelper_automation.py, linkedhelper_ghl_bridge.py, linkedhelper_gsheets_sync.py, supabase_to_linkedhelper.py, egp_a4_linkedhelper_500_build.py
feedback_persistent_autonomous_off_hours_operation_canonical_2026_05_18.md (50/day, 50-100/day caps, off-hours)
feedback_job_tracker_dup_handling_strikethrough_yellow_2026_05_18.md (dup-handling canonical)
feedback_clarissa_distinct_persona_local_llm_large_2026_05_15.md (Cody persona)

10. Persona Attribution

Betty (chair, approval gates)
Willemena (PPMO-HQ + PPMO-LH carry, lead-list and conversation owner)
Sade (Playwright executor, LH API wrapper, hardening)
Gemma (telemetry pane)
Vesta (auth, scheduling integration)
Clarise (counters, audit, ship report)
Claudia (intent classifier for LH directives)
Cody (the executor)
Sasha (Skeptic, template and new-campaign veto)