feature

name	feature
description	Start next feature — full workflow from issue triage through spec-driven development to shipping.
argument-hint	Optionally specify an issue number or feature description
compatibility	Requires spec-kit project structure with .specify/ directory
metadata	{"author":"jabelk","source":"claude-code-template"}
user-invocable	true

Start Next Feature

Full workflow from issue triage to implementation. Do NOT skip steps.

Phase 1: Orient (always do this first)

1. Read CLAUDE.md and .specify/memory/constitution.md to refresh context
2. Read MEMORY.md for project status, user preferences, and active work
3. Run gh issue list --state open --limit 30 to see the current backlog
4. Run git log --oneline -10 to see recent work
5. Present a summary: "Here's where we are, here's what's open, here's what I'd recommend next"
6. Wait for user input on which issue(s) to tackle

Phase 2: Research (do NOT skip this)

Before specifying or planning, research online:

• Are there better approaches than the obvious one?
• What do competitors/similar projects do?
• Any libraries, APIs, or patterns that would save time?
• Check if the approach has known pitfalls

Present findings to user. This step has repeatedly prevented bad architectural decisions.

Phase 3: Specify

1. Run /speckit-specify with the feature description
2. This creates the branch and spec automatically
3. Review for plot holes, missing edge cases, unclear requirements
4. Wait for user feedback before proceeding

Phase 4: Plan + Tasks

1. Run /speckit-plan — research decisions, data model, architecture
2. Run /speckit-tasks — task breakdown with dependencies
3. If anything needs clarification, run /speckit-clarify
4. If quality check needed, run /speckit-analyze

Phase 4.5: Review (do NOT skip this)

1. Run /review-plan — sends spec + plan to OpenAI gpt-5.3-codex and Gemini 2.5 Pro for independent peer review
2. The script outputs raw text from each model — you must read the output and interpret the GREEN/YELLOW/RED ratings manually
3. Present all three reviews to the user with a summary
4. If any RED: strongly recommend fixes; proceed only with explicit user approval
5. If any YELLOW: review with user and decide whether to fix first
6. If all providers are SKIPPED (missing API keys): flag to user, perform manual architectural review as fallback
7. Wait for user approval before moving to Phase 5

Phase 5: Implement

1. Run /speckit-implement — execute tasks phase by phase
2. TDD where appropriate: write failing test first, then fix, then verify
3. VERIFY BEFORE SHIPPING: For any new library integration, read the actual installed source code and run a minimal working example BEFORE writing the full implementation
4. Clean up tests, docs, and code as you go — don't leave debt
5. Verify file-by-file after any bulk operations
6. Run lint and test commands after each phase

Phase 6: Ship

1. Self-review the diff for: shell pipeline safety (pipefail), API keys in URLs, secrets/PII in external calls, doc consistency
2. Run tests — all must pass
3. git diff --stat — review, no unintended changes
4. Create PR against main
5. Address any CI/review findings before merging
6. Merge when approved
7. Verify deployment