Your LinkedIn agency, in one folder.
Think of it like this: a B2B LinkedIn agency has departments — voice extraction, content engine, outbound DMs, fulfillment, and a high-ticket program for the founders who outgrow the retainer. Each one has standards, scripts, and instincts the founder carries in their head.
LinkedIn Agency is that whole org chart, written down as a folder of plain markdown files.
Open the folder. Hand it to any AI agent — Claude, ChatGPT, Cursor, OpenClaw. The agent reads SYSTEM.md, learns your agency, learns your client, and starts running a department.
You stay in the loop where your judgement counts. Voice drift gets caught the week it starts, not the month after a client churns.
| Step | What happens |
|---|---|
| 01 | Clone the folder |
| 02 | Fill in company.yaml with your agency — voice patterns, ICP, retainer offer, program offer |
| 03 | Hand the folder to any agent. It reads SYSTEM.md, becomes a LinkedIn agency operator, and starts shipping |
| Without | With |
|---|---|
| The ghostwriter quits at month 6. Client posts go off-voice for two weeks before anyone notices. | Voice is encoded per client. Every post passes a drift check before it ships. |
| The DM operator drifts off-script by month 3. Inbound replies get classified by gut feel. | A 4-class intent taxonomy routes every reply. Booking-gate qualification runs every day. |
| Client-of-the-month posts read like last month's client. | Every client has their own voice profile, their own hook library, their own posting calendar. |
| You spend Sundays writing your own LinkedIn while writing your clients' too. | Your founder posts and your client posts are different files. Both stay in your voice. |
| The high-ticket program runs on Slack, Notion, three Loom videos, and hope. | Curriculum, cohort cadence, first-win triggers, and retention pulses live in the same workspace. |
git clone https://github.com/Heuresis/LinkedIn-Agency.git your-agency
cd your-agency
./boot.shBoot prints the workspace credentials and runtime install options. Then install into your runtime of choice:
./scripts/install.sh --tool claude-codeEleven runtimes supported. See Multi-tool integrations.
Fill in company.yaml with your agency context. Then ask for what you need:
/extract-founder-voice a voice profile (yours or a client's)
/build-icp an ideal customer profile
/design-retainer-offer a DFY service offer document
/design-program-offer a high-ticket program offer document
/build-content-engine pillars, hook library, calendar
/build-outbound-engine DM intent taxonomy + per-class sequences
/ghostwrite-client-post a client-bound post (with voice-drift check)
/run-client-dm-ops a daily client DM operations cycle
/voice-drift-detector a weekly per-client voice audit
Full setup walkthrough: Quickstart · 30 minutes.
Every Heuresis workspace is the same shape. A boot layer reads your agency context, then activates the org-chart, skill outputs, methodology brain, and trigger manifest. Eleven runtime integrations ship — see Multi-tool integrations.
┌──────────────────────────────────────────────────────────────┐
│ THE ENCODED WORKSPACE │
│ │
│ ┌───────────┐ ┌───────────┐ ┌───────────┐ ┌───────────┐ │
│ │ SYSTEM.md │ │ ENCODING │ │ INVARIANTS│ │ company │ │
│ │ boot file │ │ schema │ │ rules │ │ context │ │
│ └───────────┘ └───────────┘ └───────────┘ └───────────┘ │
│ │
│ ┌───────────┐ ┌───────────┐ ┌───────────┐ ┌───────────┐ │
│ │ agents/ │ │ skills/ │ │ reference/│ │ triggers │ │
│ │ org chart │ │ outputs │ │ brain │ │ manifest │ │
│ └───────────┘ └───────────┘ └───────────┘ └───────────┘ │
└──────────────────────────────────────────────────────────────┘
│
▼
any agent runtime that reads files — 11 ship
linkedin-agency/
│
├── README.md ← the pitch
├── SYSTEM.md ← boot file · any AI becomes a LinkedIn agency operator
├── INVARIANTS.md ← 14 NEVERs + 14 ALWAYSes (sacred rules)
├── ENCODING.md ← 6-profile schema (recursive: agency + per-client)
├── company.yaml ← YOUR agency · fill once
│
├── agents/ ← 32 specialists · an org chart of a real agency
│ agency-director top orchestrator
│ foundations-head voice · ICP · positioning · offer
│ marketing-head content engine · ghostwriting · virality
│ acquisition-head outbound DMs · application gate · sales calls
│ fulfillment-head DFY delivery · voice-drift audit · client reports
│ program-head curriculum · cohort cadence · student success
│
├── skills/ ← 30 capabilities · each produces one asset
│ │
│ │ FOUNDATIONS
│ ├── /extract-founder-voice Voice profile (yours or a client's)
│ ├── /build-icp Ideal Customer Profile
│ ├── /design-retainer-offer DFY service offer document
│ └── /design-program-offer High-ticket program offer document
│ │
│ │ MARKETING
│ ├── /build-content-engine Pillars, hook library, calendar
│ ├── /write-linkedin-post Founder post (your own LinkedIn)
│ ├── /ghostwrite-client-post Client-bound post (with voice-drift check)
│ ├── /build-hook-library 50+ hook templates per voice
│ ├── /build-posting-calendar 30-day editorial calendar
│ └── /score-hook Hook scorecard before publishing
│ │
│ │ ACQUISITION
│ ├── /build-outbound-engine DM intent taxonomy + per-class sequence
│ ├── /write-dm-sequence 4-class DM sequences
│ ├── /classify-dm-intent Auto-route inbound replies
│ ├── /build-qualification-thread Multi-message qualification flow
│ ├── /build-application Application with hard-DQ logic
│ ├── /triage-application Score + sort applications
│ ├── /source-lead-list ICP-precise, intent-scored prospect list
│ ├── /score-lead-intent Pre-DM intent score per prospect
│ ├── /build-call-prep Discovery call prep doc
│ ├── /summarize-discovery-call Post-call summary + follow-up draft
│ └── /no-show-recovery Re-engagement after a no-show
│ │
│ │ FULFILLMENT
│ ├── /onboard-client First-30-days client onboarding
│ ├── /voice-drift-detector Weekly per-client voice audit
│ ├── /post-approval-tracker Client approval workflow tracker
│ ├── /run-client-dm-ops Daily VA-driven client DM cycle
│ ├── /build-client-report Monthly client report
│ └── /build-attribution-trace Lead-to-revenue trace per client
│ │
│ │ PROGRAM
│ ├── /onboard-program-student Student onboarding (cohort start)
│ ├── /program-cohort-pulse Weekly cohort health check
│ ├── /build-curriculum Curriculum module
│ ├── /build-first-win-trigger Engineer the student's first measurable win
│ └── /build-retention-pulse Retention check (program-side)
│
└── reference/ ← the brain that makes skills smart
├── frameworks/ ICP, voice, content, DM, offer, fulfillment frameworks
├── operators/ Anonymized archetype playbooks
├── platforms/ LinkedIn, X, IG, YouTube, email playbooks
├── playbooks/ Multi-step playbook (24-month arc)
├── swipe-file/ Annotated high-performer LinkedIn posts
├── templates/ Output templates (proposal, onboarding, report)
└── verticals/ Vertical adaptations (SaaS, services, e-com, info-product)
Each file is plain text. Each folder is owned by you. Nothing is locked behind an app.
- Voice profile — extracted in week 1, ship-grade
- ICP — per-client with verbatim voice-of-customer language
- Content engine — pillars, hook library, posting calendar
- DM engine — 4-class intent taxonomy, per-intent sequences, booking-gate qualification
- Lead list — ICP-precise, intent-scored, prioritized
- Discovery call prep — research-deep, objection pre-framed
- Retainer proposal — transformation hypothesis, named mechanism, value stack
- Monthly report — signal-dense, attribution-clean
- Voice-drift audit — weekly per client
Every action leaves a receipt. Every post passes the drift check. Every client retains a profile that gets sharper every cycle.
Wire the workspace into a runtime with cron, webhook, and event triggers. The agents keep working without you in the room.
- Monday 09:00 (cron) —
voice-drift-detectorruns across every active client - On
lead.applied(webhook) — application response triaged, DQ logic runs, qualified leads booked - On
call.ended(event) — call recording transcribed, summary generated, follow-up DM drafted, CRM updated - Daily 06:00 (cron) — content engine queues today's agency post + today's per-client posts
- Continuous (event) — DM ops run intent classification on inbound replies, route to the right sequence
Triggers are declared in paperclip.manifest.yaml. Wire them to your scheduler and the workspace operates as a 24/7 LinkedIn agency.
5 agent-first departments. 32 agents. 30 skills.
- Foundations — voice extraction · ICP build · positioning · offer architecture
- Marketing — content engine · hook library · per-client ghostwriting · cross-platform repurposing
- Acquisition — outbound DM engine · 4-class intent taxonomy · application gate · call prep · closing · follow-up
- Fulfillment — account management · ghostwriting protocol · client DM ops · lead-list sourcing · monthly reporting · voice-drift detection
- Program — curriculum design · mentorship cadence · community rituals · first-win engineering · retention pulse
Each department is a full department rebuilt as agent-first — agents that carry your judgement, read signals, and make decisions.
The workspace is files. Files run anywhere that reads files. The Agency ships conversion + install scripts so the same agents work across every major agentic coding tool — no vendor lock, no rewrite.
| Tool | Install | Destination |
|---|---|---|
| Claude Code | ./scripts/install.sh --tool claude-code |
~/.claude/agents/heuresis-linkedin-agency/ |
| GitHub Copilot | ./scripts/install.sh --tool copilot |
~/.github/agents/ + ~/.copilot/agents/ |
| Antigravity (Gemini) | ./scripts/install.sh --tool antigravity |
~/.gemini/antigravity/skills/ |
| Gemini CLI | ./scripts/install.sh --tool gemini-cli |
~/.gemini/extensions/heuresis-linkedin-agency/ |
| OpenCode | ./scripts/install.sh --tool opencode |
./.opencode/agents/ |
| Cursor | ./scripts/install.sh --tool cursor |
./.cursor/rules/ |
| Aider | ./scripts/install.sh --tool aider |
./CONVENTIONS.md |
| Windsurf | ./scripts/install.sh --tool windsurf |
./.windsurfrules |
| OpenClaw | ./scripts/install.sh --tool openclaw |
~/.openclaw/heuresis-linkedin-agency/ |
| Qwen Code | ./scripts/install.sh --tool qwen |
./.qwen/agents/ |
| Kimi Code | ./scripts/install.sh --tool kimi |
~/.config/kimi/agents/ |
Five tools (Antigravity, Gemini CLI, OpenClaw, Cursor, Kimi) require format conversion — the install script handles it transparently. To inspect the converted artifacts before deploying:
./scripts/convert.sh --tool <tool-name>
ls build/<tool-name>/Every tool has its own README with activation patterns, customization notes, and uninstall steps. Full list: integrations/README.md.
Runtime-swappable. Your workspace is the asset. The runtime is replaceable.
What every Heuresis workspace ships with, regardless of which runtime you point at it.
| Property | What it does |
|---|---|
| Plain-text workspace | Every agent, skill, and framework is markdown + YAML. Read it, fork it, version it, diff it. |
| Runtime-portable | The same files run across every major agentic tool. Eleven integrations ship — see Multi-tool integrations. |
| Goal ancestry | Every skill carries the agency's positioning, ICP, offer, and ghostwriting voice as live context. Agents act on the why, not just the what. |
| Encoded judgement | Agents carry the operator's voice extraction, qualification bar, and ghostwriting standards. The workspace replicates the operator's judgement, not their task list. |
| Receipt trail | Every cycle leaves an artefact — a post, a DM thread, an application score, a voice-drift report. Auditable end-to-end. |
| Compounding library | Wins, hook patterns, objection captures, and per-client benchmarks feed back into the workspace. Every cycle sharpens the next. |
| Trigger-ready | Cron, webhook, and event triggers declared in a manifest. Wire to any scheduler — daily client posts, weekly voice-drift audits, inbound DM routing — and the workspace operates while you sleep. |
| Owned outright | MIT-licensed. Fork it, host it, run it forever. Yours. |
Every founder-led LinkedIn agency hits the same wall: the founder IS every department.
The ghostwriter quits at month 6. The DM operator drifts off-script by month 3. Client-of-the-month posts read like last month's client. The high-ticket program runs on Slack, Notion, three Loom videos, and hope.
Encoding changes the shape of the week. Your judgement — voice extraction, ICP build, DM intent classification, ghostwriting standards — gets written into agents that run each department on your behalf. The founder stays. The judgement scales. The agency compounds.
Every cycle, each department gets sharper. The gap between your agency and every competitor operating off memory widens.
This is one template in the library. More shipping, vertical by vertical. Every outcome claim we publish traces to a real deployment with a real operator. Thesis, method, and source go public on ideas. Receipts wait their turn.
- ✅ 5 departments · 32 agents · 30 skills shipped
- ✅ Voice extraction · DM intent taxonomy · application gate
- ✅ Voice-drift detection (weekly per-client audit)
- ✅ Vertical adaptations: SaaS · services · e-com · info-product
- ✅ Eleven runtime integrations: Claude Code · Copilot · Gemini · Cursor · Aider · Windsurf · OpenClaw · Qwen · Kimi · OpenCode · Antigravity
- ✅ Trigger manifest: cron · webhook · event
- ⚪ Multi-client voice library + drift telemetry dashboard
- ⚪ Inbound classifier auto-routing to setter agents
- ⚪ Per-client KPI dashboards
- ⚪ Cohort-aware program operations
- ⚪ Outbound lead-source benchmarking
- ⚪ Heuresis Cloud — managed hosting for the workspace
- Quickstart — setup in 30 minutes
- Architecture — how the folder is built
- Skill Authoring — write your own agents and skills
MIT. Free forever.
Built by Syed Hussain at Heuresis.