Growth Acquisition for DeepSeek Harness

English | 中文

dsh-growth 是一个本地优先的 DeepSeek Harness 插件，用于把 Markdown、CSV 和 JSONL 中的增长资料与业务数据，转化为可解释的增长诊断、落地方案和执行 SOP。

dsh-growth is a local-first DeepSeek Harness bundle for evidence-backed user growth and customer acquisition analysis, actionable plans and execution SOPs.

Public six-plugin collaboration contract: SUITE.md.

协作可靠与归因边界

growth_artifact_review 校验增长工件的 ID、内容指纹和有效期；growth_attribution_review 将 GEO/内容测量计划与实际指标匹配，只对具备 baseline 的观测进行比较，不自动宣称因果。

growth_artifact_review validates growth artifacts, fingerprints and freshness; growth_attribution_review matches content measurement plans to observed metrics and compares only baseline-backed observations without claiming causality.

It covers AARRR funnels, activation, retention cohorts, referral loops, MRR bridges, CAC/LTV/payback, HADI experiments, RICE prioritization and WBR/MBR reports for Markdown, CSV and JSONL data.

Plugin Positioning and Collaboration Navigation

dsh-growth is the growth operating layer in the six-plugin system. It connects acquisition, activation, retention, revenue and experiment data to answer “what changed, why did it change and what should happen next?”

• Owns: AARRR funnels, activation, retention cohorts, referrals, MRR, CAC/LTV/Payback, growth diagnosis, HADI experiments, prioritization and WBR/MBR/QBR.
• Inputs: Product/PMF event definitions from [dsh-product](../dsh-product/README.md), pricing and margin rules from [dsh-business](../dsh-business/README.md), sales pipeline data from [dsh-sales](../dsh-sales/README.md), and content/search acquisition signals from [dsh-geo](../dsh-geo/README.md).
• Outputs: Traceable metric diagnoses, funnel/retention/revenue analyses, experiment cards, operating reviews and next actions for product, commercial strategy, sales and opportunity discovery.
• Does not own: Product definition, commercial pricing, sales follow-up or website engineering. It analyzes and plans growth; it does not treat unvalidated metrics as conclusions.

Positioning Architecture: Commercial Strategy Layer + Four-Stage Core Flow

The six plugins work together to turn a real demand signal into a deliverable product, reach target customers through marketing, and use monetization results to drive product iteration or discover new opportunities.

flowchart TB
    S["dsh-business<br/>Commercial strategy layer<br/>Value · model · pricing · profit"]

    subgraph MAIN["Four-stage core flow"]
        direction LR
        A["1. Demand<br/>dsh-idea"] --> B["2. Product<br/>dsh-product"]
        B --> C["3. Marketing<br/>dsh-geo + dsh-growth"]
        C --> D["4. Monetization execution<br/>dsh-sales"]
    end

    S -. "Sets commercial direction and guardrails" .-> A
    D --> R["Feedback<br/>Deals · renewals · revenue · cost"]
    R -->|Product iteration| B
    R -->|New discovery| A

    classDef strategy fill:#FFF4D6,stroke:#B7791F,color:#5C4500
    classDef stage fill:#E8F1FF,stroke:#3366CC,color:#173A7A
    classDef feedback fill:#E8F7EE,stroke:#2F855A,color:#1C4532
    class S strategy
    class A,B,C,D stage
    class R feedback

This plugin covers growth measurement and experimentation in the marketing stage: it puts the discoverability from [dsh-geo](../dsh-geo/README.md), product behavior and closes from [dsh-sales](../dsh-sales/README.md) into one measurement and experimentation framework. [dsh-business](../dsh-business/README.md) provides commercial goals and economic boundaries; when monetization results signal a problem, the evidence feeds [dsh-product](../dsh-product/README.md) for product iteration or [dsh-idea](../dsh-idea/README.md) for new opportunity discovery.

Plugin Navigation

Recommended Handoffs

User and company pain points

Growth work often breaks down in the gap between data, decisions and execution:

Application scenarios

Included tools

The optional growth-acquisition-execution skill adds external channel resources and a controlled directory-submission planning SOP. It is intentionally separate from deterministic metric tools because live site rules, browser state and human verification change over time. It only qualifies resources, prepares plans and handoffs; it does not browse, create accounts, fill forms or submit to external sites.

The optional growth-ai-discoverability skill adds an AI Search / Discoverability Readiness method for deciding which crawlability, structured-data, content-trust and commerce-feed checks apply to a business. It produces a readiness matrix and implementation plan; it does not scan or modify websites and does not depend on GEO-PRO.

The optional growth-strategy-planning skill routes classic growth methods such as Value Proposition Canvas, Lean Canvas, Bullseye channels, Activation Events, Churn/Win-back, Opportunity Solution Trees, user interviews, referral loops, market sizing, pricing research, B2B revenue funnels and Growth Accounting into small, evidence-backed planning artifacts. It does not add live execution or replace the deterministic metric tools.

Quick start

Zero-threshold path

You do not need to know the tool names, AARRR definitions, field mappings or which dataset to open first. You only need:

1. A business question, such as “why did activation fall?” or “which acquisition channel should we scale?” 2. A configured local growth root, or a file path if you already know the relevant file. 3. Permission to read local data; the plugin does not upload your vault.

For a new project, start with a readiness check:

Run a growth onboarding check for my configured root.
Tell me what is ready, partial, missing and not supported, and give me only the top two gaps to fix next. Do not write files.

It checks growth notes and local CSV/JSON/JSONL datasets without returning raw user rows. It also shows which classic methods are detected in the project, which are only available as audits or templates, and which require an external system. If you already know the goal and want analysis immediately, use growth_review instead.

The shortest first review request is:

Run a growth review for the goal "improve activation" using the best available data under my configured root.
Show me which files you selected, what is missing, the biggest bottleneck and the next check. Do not write files.

If no paths are supplied, growth_review scans the configured root, profiles supported CSV/JSON/JSONL files, selects the most analysis-ready event and MRR sources, and explains the selection. If there are several candidates, confirm the selected files before making a budget or product decision.

Install the plugin into a DeepSeek Harness profile:

npx --yes @deepseek-ai/dsh plugin --profile growth add dsh-growth
npx --yes @deepseek-ai/dsh --profile growth --dump-config

If the Harness host is not running yet, start its Web UI first:

npx --yes @deepseek-ai/dsh web

Then open http://127.0.0.1:3080 in your browser and use the growth profile. DeepSeek Harness's official repository documents this Web UI entry point; the host is in developer preview, so its setup commands may evolve.

If dsh is already on your PATH, the equivalent short form is dsh plugin --profile growth add dsh-growth.

Configure the plugin through the host. A minimal configuration is:

defaultRoot: "<your-local-growth-root>"
reportDir: ".dsh-growth/reports"
defaultCurrency: "CNY"
defaultTimezone: "Asia/Shanghai"

Then use the tools from the conversation. Typical requests are:

Run a growth review for the goal "improve activation" using the best available data under my configured root; show which files you selected and what is missing.
Run a growth review for the goal "improve activation" using events.csv; tell me what is missing before making a recommendation.
Audit growth-plan.md for PMF, North Star, AARRR metrics and evidence gaps.
Analyze events.csv as an AARRR funnel and compare channel and segment performance.
Analyze mrr.csv for MRR Bridge, NRR, CAC, LTV and Payback using a gross margin of 0.8.
Turn the largest activation bottleneck into a HADI experiment and score it with RICE.
Generate this week's WBR as Markdown; do not write a file yet.

The recommended conversation flow

Use these six requests in order when you are new to the plugin:

1. Run a growth onboarding check; tell me what is ready, partial, missing and not supported.
2. Review the goal "improve activation" and tell me what is missing before giving a conclusion.
3. Break down the bottleneck by channel and segment; separate evidence from hypotheses.
4. Turn the highest-leverage hypothesis into a HADI experiment with a primary metric and guardrails.
5. Score the experiment with RICE and show which inputs are estimates rather than observed facts.
6. Generate this week's WBR; preview only and do not write a file.

You can replace the goal and the file names without changing the workflow. Advanced users may call the individual tools directly, but that is optional.

The operating SOP

Treat the workflow as six decision gates, not as a list of tools:

If a gate fails, ask for the smallest missing input and stop there. Do not invent a metric, treat correlation as causality, or write a report before preview and confirmation.

The actual analysis tools are registered by the DeepSeek Harness/Cordis host. A Codex plugin installation may load the growth-operator guidance without exposing the Cordis tools themselves; verify that growth_onboarding appears in the host's callable tool list before treating the installation as operational.

Tool results use a stable envelope with ok, data, warnings, assumptions, lineage and nextActions. Read warnings and lineage before using a number in a decision.

External acquisition planning and submission SOP

For product directories or discovery channels, invoke the optional planning skill with a clear business outcome:

Use $growth-acquisition-execution to find relevant product-directory channels for our target market.
Start with the local candidate resource, define the live recheck and terms checklist, run the quality gate,
prepare no more than 10 pilot site handoffs, and do not browse or submit anything.
Record the evidence fields and explain how future referral visits and activation would connect back to the growth review.

The skill distinguishes quality-pilot planning from pre-approved batch planning. It does not perform live external actions, bypass CAPTCHA or verification, invent product facts, or treat submission count as growth impact.

AI search discoverability planning

For AI search, product discovery or LLM-readiness questions, use the optional planning skill:

Use $growth-ai-discoverability to assess whether our product is ready for AI search and product discovery.
First classify the business model and outcome, then build a readiness matrix for crawlability,
machine-readable facts, content trust and commerce-feed applicability. Output gaps, owners,
acceptance criteria and validation metrics only; do not scan or modify the website.

It separates general search fundamentals from platform-specific advice, marks unknowns as needs-external-validation, and does not promise ranking, indexing, citations or conversion.

Classic growth method planning

When the question is about choosing a framework or turning a vague growth problem into a plan, use the optional planning skill:

Use $growth-strategy-planning for our current growth problem.
Choose the smallest useful classic method, separate observed evidence from assumptions,
produce the planning artifact, connect it to one primary metric and guardrails,
and turn the riskiest assumption into a HADI experiment. Do not invent missing inputs.

It routes the request instead of stacking frameworks: Value Proposition / Lean Canvas for context, Bullseye for channels, Aha Moment for activation, Churn/Win-back for retention, Opportunity Solution Tree for diagnosis-to-solution, and pricing/B2B/Growth Accounting methods only when the business model requires them.

Input conventions

Event data should use user_id, event and timestamp, with optional channel, segment, plan, revenue and currency fields. MRR data should use period, type, amount, customer_id, active_customers and spend. Supported movement types are new, expansion, reactivation, contraction, churn and churned. The goal-oriented review recognizes common English and Chinese event values such as signup / 注册, activated / 激活, active / 活跃, invited / 邀请 and paid / 付费.

For the first review, eventPath and economicsPath can be omitted. growth_review scans the configured local root, selects the most analysis-ready event and MRR files, and records the selected sources in assumptions, warnings and lineage. If more than one file is suitable, confirm the selection before using the result for a budget or product decision.

Minimal data examples

An event file can start with only these three fields:

user_id,event,timestamp
u001,signup,2026-08-01T09:00:00Z
u001,activated,2026-08-01T09:20:00Z
u001,active,2026-08-08T09:20:00Z

For MRR and acquisition cost analysis, add the fields you actually have:

period,type,amount,customer_id,active_customers,spend,currency
2026-08,new,1000,c001,20,5000,CNY
2026-08,expansion,200,c002,20,,CNY
2026-08,churned,100,c003,20,,CNY

Do not fabricate missing columns. The plugin will keep dependent metrics unavailable and explain what needs to be added.

How to read a result

Every tool returns the same outer structure:

…