# Feature Pipeline

> Full feature workflow — deep-interview → prd → consensus-plan → tdd → code-review → security-gate

- Skill: `andrem-sec/feature-pipeline` (Agent Skill)
- Install (CLI): `npx skillmds@latest add andrem-sec/feature-pipeline`
- Raw SKILL.md: https://api.skillmd.com/api/skills/andrem-sec/feature-pipeline/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Product & Planning
- Author: andrem-sec (https://skillmd.com/u/andrem-sec)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/andrem-sec/feature-pipeline

---


# Feature Pipeline Skill

The complete workflow for building a new feature — from unclear idea to reviewed, secure, committed code.

## What This Skill Does

Orchestrates the full sequence of skills required to take a feature from "we should do X" to "X is done and verified." Each stage gates on the previous one. No stage can be skipped without explicit acknowledgment.

## Stage Selection

Not every feature needs every stage at full depth. Assess on intake:

| Feature Type | Explore | Specify | Plan | Implement | Review | Security |
|-------------|---------|---------|------|-----------|--------|---------|
| Small, clear | skip | prd (light) | plan-first | tdd | code-review | gate |
| Medium, some unknowns | deep-interview | prd | plan-first | tdd | code-review | gate |
| Large, complex | deep-interview | prd | consensus-plan | tdd | code-review | gate |
| Security-sensitive | deep-interview | prd | consensus-plan | tdd | code-review + security-reviewer | gate |

State which stages will run and why before starting.

## Gate Protocol

Every stage ends with a gate. The gate is explicit user confirmation before the next stage begins.

Gate format:
```
[Stage name] complete.
Output: [one-line summary of what was produced]
Gate: Proceed to [next stage]? (yes / modify / stop)
```

Do not proceed without "yes" or equivalent. "Looks good" is yes. Silence is not.

## Checkpoint Rule

During the Implement stage: checkpoint every 5 sequential steps. The pipeline does not override the 5-step checkpoint rule.

## State Persistence

At every gate boundary, write a state entry to `context/learnings.md`:
```
[date] state — Task: [feature name] | Step: [current stage] | Done: [completed stages] | Next: [next stage]
```

This enables `resume` to pick up if the pipeline is interrupted.

## Handoff Between Stages

Each stage produces an artifact that the next stage consumes:

| Stage | Produces | Next stage consumes |
|-------|---------|-------------------|
| deep-interview | Understood problem (interview summary) | prd uses findings as context |
| prd | Confirmed acceptance criteria | plan references criteria by number |
| consensus-plan / plan-first | Approved plan | tdd implements plan steps in order |
| tdd | Passing tests + implementation + test run evidence (pass/fail counts) | code-review scopes to changed files |
| code-review | APPROVE verdict | security-gate scopes to same files |
| security-gate | PASS verdict | wrap-up closes out |

## Anti-Patterns

Do not start implementation before the PRD is confirmed. The most common cause of rework.

Do not run code-review and security-gate in the same session that wrote the code. Spawn isolated agents.

Do not skip the close step. Learnings from a completed feature are high-value for future features.

## Mandatory Checklist

1. Verify stage selection was stated and justified before starting
2. Verify each stage produced its expected artifact before the gate was opened
3. Verify state was written to learnings.md at every gate boundary
4. Verify no stage was skipped without explicit acknowledgment
5. Verify code-review and security-gate used isolated agents
6. Verify wrap-up was run at close