From pde
Product Delivery Engine — automate DiffLab's 6-stage product pipeline (INTAKE → FILTER → VALIDATE → BUILD → LAUNCH → REVENUE) with specification-driven workflows and human approval gates. Separates high-level product aims (behavior, evaluation) from implementation details (numbered implementation folders). Use when: (1) evaluating a new product idea or market signal, (2) advancing a product through pipeline stages, (3) creating validation landing pages and survey funnels, (4) generating stage artifacts, (5) checking product status, (6) decommissioning a product. Triggers on 'new product idea', 'product brief', 'behavior spec', 'evaluate product', 'validate product', 'landing page', 'survey funnel', 'build spec', 'launch plan', 'revenue report', 'product pipeline', 'advance stage', 'decommission product'.
How this skill is triggered — by the user, by Claude, or both
Slash command
/pde:pdeThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Specification-driven product pipeline. High-level aims (behavior + evaluation) are separated from implementation details (numbered implementation folders). Each stage produces markdown artifacts committed to the configured PDE location (see Configuration) with human approval before advancing.
assets/templates/architecture.mdassets/templates/behavior.mdassets/templates/build-spec.mdassets/templates/business-model-analysis.mdassets/templates/evaluation.mdassets/templates/icp-analysis.mdassets/templates/landing-page.htmlassets/templates/launch-plan.mdassets/templates/monetization-analysis.mdassets/templates/revenue-report.mdassets/templates/survey-funnel.htmlassets/templates/tests.mdassets/templates/validation-plan.mdassets/templates/validation-results.mdreferences/pipeline.mdreferences/validation.mdSpecification-driven product pipeline. High-level aims (behavior + evaluation) are separated from implementation details (numbered implementation folders). Each stage produces markdown artifacts committed to the configured PDE location (see Configuration) with human approval before advancing.
Before executing any pipeline stage, resolve the PDE storage location:
.project/settings.local in the current workspace root. Look for a line matching pde_location = <value>.claude/settings.local in the current workspace root. Look for the same key.owner/repo) or a local path (e.g., ./pde or /absolute/path/to/pde)."Store the resolved value as {pde-location} for all operations in this skill.
Detecting location type:
{pde-location} matches the pattern owner/repo (exactly one /, no leading /, ., or ~), treat it as a remote GitHub repository.behavior.md from assets/templates/behavior.mdproducts/{product-name}/behavior.md and present to userbehavior.md and evaluation.md for high-level contextScan products/{product-name}/ and apply the first matching rule (top to bottom):
DECOMMISSIONED.md exists → Product decommissioned — stoprevenue-report.md exists → Stage 6 activeimplementation-NN/launch-plan.md exists (highest NN) → Stage 5 in progressimplementation-NN/build-spec.md exists (highest NN) → Stage 4 in progressvalidation-results.md exists → Stage 3 complete — ready for Buildvalidation-plan.md exists → Stage 3a in progress — awaiting validation dataevaluation.md exists → Stage 2 complete — ready for Validateicp-analysis.md, monetization-analysis.md, business-model-analysis.md) → Analysis complete — ready for Filterbehavior.md exists → Stage 1 complete — launch parallel analysisEach product has two layers:
High-level (product-wide, implementation-independent):
behavior.md — Desired behavior and value propositionevaluation.md — Success criteria and pass/fail thresholds for every stagevalidation-plan.md / validation-results.md — Market validation datarevenue-report.md — Revenue metrics and scale decisionsImplementation-specific (in numbered folders):
implementation-01/architecture.md — Tech stack, infrastructure decisionsimplementation-01/build-spec.md — MVP features, timeline, team allocationimplementation-01/tests.md — Behavioral tests derived from behavior.mdimplementation-01/launch-plan.md — Soft launch strategy, metrics setupimplementation-01/assets/ — Implementation-specific copies of launch assets (originals live in products/{product-name}/assets/)Always work in the current (highest-numbered) implementation folder unless the user explicitly says they want a new implementation.
To start a new implementation: create implementation-NN/ where NN is the next number. New implementations inherit the same behavior.md and evaluation.md — only the implementation approach changes.
All product artifacts go in products/{product-name}/ within {pde-location}. The commands differ based on whether the location is a remote GitHub repository or a local path.
Use the gh CLI for all repo operations — do not clone the repo locally.
gh api repos/{pde-location}/contents/products/{product-name}/{file}.md \
--jq '.content' | base64 -d
# Write a new file (no SHA needed)
gh api repos/{pde-location}/contents/products/{product-name}/{file}.md \
--method PUT \
--field message="{Stage}: {product-name} — {brief description}" \
--field content="$(base64 -i /path/to/local/file.md)"
# Update an existing file (SHA required)
SHA=$(gh api repos/{pde-location}/contents/products/{product-name}/{file}.md --jq '.sha')
gh api repos/{pde-location}/contents/products/{product-name}/{file}.md \
--method PUT \
--field message="{Stage}: {product-name} — {brief description}" \
--field content="$(base64 -i /path/to/local/file.md)" \
--field sha="$SHA"
gh api repos/{pde-location}/contents/products/{product-name} --jq '.[].name'
Use standard file system operations.
cat {pde-location}/products/{product-name}/{file}.md
mkdir -p {pde-location}/products/{product-name}
cp /path/to/local/file.md {pde-location}/products/{product-name}/{file}.md
ls {pde-location}/products/{product-name}/
products/{product-name}/ within the current workspaceUse kebab-case: ai-code-reviewer, spec-validator, landing-gen
| Stage | File | Layer |
|---|---|---|
| 1 Intake | behavior.md | High-level |
| 1+ ICP Analysis | icp-analysis.md | High-level |
| 1+ Monetization | monetization-analysis.md | High-level |
| 1+ Business Model | business-model-analysis.md | High-level |
| 2 Filter | evaluation.md | High-level |
| 3a Validate | validation-plan.md | High-level |
| 3b Validate | validation-results.md | High-level |
| 4 Build | implementation-NN/architecture.md | Implementation |
| 4 Build | implementation-NN/build-spec.md | Implementation |
| 4 Build | implementation-NN/tests.md | Implementation |
| 5 Launch | implementation-NN/launch-plan.md | Implementation |
| 6 Revenue | revenue-report.md | High-level |
| Decommissioned | DECOMMISSIONED.md | High-level |
Input: Market signals, customer conversations, training insights, AI trends
Output: products/{product-name}/behavior.md
assets/templates/behavior.mdproducts/{product-name}/behavior.md and present to userAfter the user approves behavior.md, launch 3 parallel analysis subagents. Send all 3 Task tool calls in a single message so they run concurrently.
Before launching, verify the companion skills exist (sibling directories within this plugin):
../icp-analysis/SKILL.md../monetization-analysis/SKILL.md../business-model-analysis/SKILL.mdIf any skill file is missing, warn the user and skip that analysis.
Launch these 3 Task calls in one message.
For each subagent prompt, include the resolved {pde-location} value and whether it is a remote repo or local path. For remote repos, subagents use gh api repos/{pde-location}/contents/... commands. For local paths, subagents use standard file system commands (cat to read, mkdir -p + cp to write).
Task call 1 — ICP Analysis:
subagent_type: general-purposedescription: "ICP analysis for {product-name}"prompt: "You are an Ideal Customer Profile analyst. The PDE location is '{pde-location}' ({remote|local}). Read the behavior spec: {read-command for products/{product-name}/behavior.md}. Find the PDE plugin directory by running: Glob for '**/plugins/pde/skills/icp-analysis/SKILL.md'. Read the ICP analysis skill instructions from that file. Read the output template from the sibling pde skill's assets: the template is at the same plugin path under skills/pde/assets/templates/icp-analysis.md. Conduct research using WebSearch, then fill in the template — replace all [Placeholder] markers with actual values. Write the result locally to products/{product-name}/icp-analysis.md, then push it to the PDE location: {write-command for products/{product-name}/icp-analysis.md with message 'Analysis: {product-name} — ICP'}"Task call 2 — Monetization Analysis:
subagent_type: general-purposedescription: "Monetization analysis for {product-name}"prompt: "You are a monetization strategy analyst. The PDE location is '{pde-location}' ({remote|local}). Read the behavior spec: {read-command for products/{product-name}/behavior.md}. Find the PDE plugin directory by running: Glob for '**/plugins/pde/skills/monetization-analysis/SKILL.md'. Read the monetization analysis skill instructions from that file. Read the output template from the sibling pde skill's assets: the template is at the same plugin path under skills/pde/assets/templates/monetization-analysis.md. Conduct research using WebSearch, then fill in the template — replace all [Placeholder] markers with actual values. Write the result locally to products/{product-name}/monetization-analysis.md, then push it to the PDE location: {write-command for products/{product-name}/monetization-analysis.md with message 'Analysis: {product-name} — monetization'}"Task call 3 — Business Model Analysis:
subagent_type: general-purposedescription: "Business model analysis for {product-name}"prompt: "You are a business model analyst. The PDE location is '{pde-location}' ({remote|local}). Read the behavior spec: {read-command for products/{product-name}/behavior.md}. Find the PDE plugin directory by running: Glob for '**/plugins/pde/skills/business-model-analysis/SKILL.md'. Read the business model analysis skill instructions from that file. Read the output template from the sibling pde skill's assets: the template is at the same plugin path under skills/pde/assets/templates/business-model-analysis.md. Conduct research using WebSearch, then fill in the template — replace all [Placeholder] markers with actual values. Write the result locally to products/{product-name}/business-model-analysis.md, then push it to the PDE location: {write-command for products/{product-name}/business-model-analysis.md with message 'Analysis: {product-name} — business model'}"After all 3 Task results return:
Input: behavior.md + all 3 analysis files
Output: products/{product-name}/evaluation.md
behavior.md, icp-analysis.md, monetization-analysis.md, and business-model-analysis.mdassets/templates/evaluation.mdAsynchronous validation via landing pages and survey funnels. See references/validation.md for the complete guide.
Output: products/{product-name}/validation-plan.md + landing page and survey assets
behavior.md and evaluation.md for contextassets/templates/validation-plan.mdevaluation.md thresholds)assets/templates/landing-page.html — customize with messaging from behavior.mdassets/templates/survey-funnel.htmlproducts/{product-name}/assets/ (shared, not implementation-specific)Output: products/{product-name}/validation-results.md
products/{product-name}/assets/ — uncomment the Stage 3b sections (solution description, pricing tiers, waitlist form) and fill in values from the behavior spec and validation dataevaluation.mdassets/templates/validation-results.mdOutput: products/{product-name}/implementation-NN/ with architecture.md, build-spec.md, tests.md
behavior.md and evaluation.md for high-level requirementsmkdir -p products/{product-name}/implementation-01/assets
assets/templates/architecture.md — define tech stack, infrastructure, key decisionsassets/templates/tests.md — derive behavioral tests from behavior.mdassets/templates/build-spec.md — define MVP features, timeline, team allocationOutput: products/{product-name}/implementation-NN/launch-plan.md
behavior.md, evaluation.md, and all implementation filesassets/templates/launch-plan.mdevaluation.md thresholdsproducts/{product-name}/assets/ into implementation-NN/assets/ for the specific launch configuration (originals stay in the shared location)Output: products/{product-name}/revenue-report.md (product-level, not per-implementation)
assets/templates/revenue-report.mdevaluation.mdWhen filling in a template:
assets/templates/[Placeholder] or [PLACEHOLDER] marker with actual values derived from research and prior artifacts<!-- Stage 3b additions -->), leave them commented until the relevant stage instructs you to uncomment themEvery stage follows this pattern:
DECOMMISSIONED.md with reasonWhen a product fails at any stage:
products/{product-name}/DECOMMISSIONED.md with:
evaluation.md failed)"DECOMMISSIONED: {product-name} — {reason}"For Intake and Filter stages, use WebSearch to gather:
If WebSearch returns no useful results for a query, try alternative search terms (broader or more specific). If research is still insufficient, note the gap in the artifact and flag it to the user — do not fabricate data.
Every training session and customer conversation is a potential Intake signal. When the user mentions insights from workshops, training, or customer calls, treat them as Stage 1 input and offer to create a behavior spec.
For detailed stage guidance:
references/pipeline.md — complete criteria, scoring models, and stage workflowsreferences/validation.md — landing page generation, survey design, analytics setup, deployment, and response analysisCreates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.
npx claudepluginhub difflabai/marketplace --plugin pde