From rafayels-engineering
End-to-end feature pipeline — from idea to merged PR with documentation updates. Orchestrates brainstorm, plan, work, review, and compound workflows.
npx claudepluginhub rafayelgardishyan/rafayels-engineering[optional: feature idea or description]re/# Feature Pipeline **Note: The current year is 2026.** Use this when dating documents. Orchestrate the full feature lifecycle: gather context, brainstorm, plan, implement in a worktree, create a PR, review, compound knowledge, and update documentation. ## Feature Description <feature_description> #$ARGUMENTS </feature_description> ## Execution Flow ### Phase -1: Select Strategy Before executing any phases, select a workflow strategy. 1. **Discover available strategies**: List all `.md` files in `references/strategies/`: 2. **Check for `--strategy` argument**: If the user passed...
/featureOrchestrates full feature development cycle: PRD with product-owner, visual style options, architecture, code, QA, docs, delivery via phased agents and gates.
/featureAutomates Rails feature/chore from Linear URL or description: creates branch, gathers context, implements/tests/lints/i18n, drafts PR, monitors CI.
/featureCreates a lightweight feature-focused PRD via guided interactive interview (5-8 rounds) optimized for new features in existing products, saving to docs/prd/{slug}.md
/featureExecutes Mermaid flowchart workflow: two Analyze-Design(D3)-Develop(G2→G2.5→G3)-Release cycles, using Task for sub-agents, AskUserQuestion at diamonds, and F5 gates.
/featureGuides through 10-stage new feature development workflow from brainstorming to code review, pausing for user confirmation at each stage.
/featureImplements features from Linear/Jira issues or prompts using minimal iteration cycles and structured workflow with resume detection, producing code increments, tests, reviews, and PRs. Accepts [issue-key-or-prompt] [--provider=linear|jira|prompt] [--silent=true|false].
Note: The current year is 2026. Use this when dating documents.
Orchestrate the full feature lifecycle: gather context, brainstorm, plan, implement in a worktree, create a PR, review, compound knowledge, and update documentation.
<feature_description> #$ARGUMENTS </feature_description>
Before executing any phases, select a workflow strategy.
Discover available strategies: List all .md files in references/strategies/:
ls references/strategies/*.md 2>/dev/null
Check for --strategy argument: If the user passed --strategy=<name>, load that strategy file directly.
If no strategy specified, ask the user: Read the "When to Use" section of each discovered strategy file. Present them via AskUserQuestion:
"Which workflow strategy should we use?"
- Full Process (Recommended) — Complete pipeline: brainstorm → plan → work → review → compound → docs
- Quick Spike — Fast prototype: skip brainstorm, lightweight plan, minimal review
- Security First — Maximum rigor: threat modeling, all security reviewers, no auto-merge
- Review Only — Audit existing code: skip implementation, run all reviewers
Load the selected strategy file and read it as natural-language guidance. Before each subsequent phase:
enabled: false for that phase → skip the phase entirelyguidance: text → follow that guidanceComposition: If the strategy has base: <other-strategy>, load that base strategy first and merge. The overlay's phase keys replace base keys per-phase; everything else inherits.
Before anything else, understand the project landscape.
<parallel_tasks>
Documentation scan: Read project documentation to understand architecture, conventions, and constraints.
find docs/ -name "*.md" -type f 2>/dev/null | head -20
cat README.md 2>/dev/null
cat CLAUDE.md 2>/dev/null
Vault research (conditional): If obsidian-adr or obsidian MCP tools are available, dispatch the vault-researcher agent in parallel with documentation scan.
Dev logs: Check recent dev logs for context on recent work and decisions.
skill: dev-log
Review the last 5 dev log entries to understand recent changes and ongoing work.
ADR search: Query Architecture Decision Records for relevant past decisions.
skill: using-adr-plugin
Run semantic searches related to the feature area. Read full ADRs for any relevant hits. Traverse the graph for connected decisions.
</parallel_tasks>
Synthesize context: Summarize key findings that will inform the feature:
If the feature description above is empty, use AskUserQuestion: "What feature would you like to implement? Describe the problem you're solving or the functionality you want to add."
Do not proceed until you have a clear feature description.
Present context summary to the user: "Based on project research, here's what I found relevant to this feature: [summary]. Does this align with your understanding? Anything to add?"
Set up an isolated workspace for this feature.
skill: git-worktree
Create a new worktree with a meaningful branch name derived from the feature description (e.g., feat/user-authentication-flow).
Important: All subsequent work happens in this worktree. Ensure you're operating in the worktree directory.
Run the brainstorming workflow to explore requirements and approaches.
/workflows:brainstorm <feature_description_with_context>
Pass the feature description along with the synthesized project context from Phase 0.
Wait for the user to refine the brainstorm. The brainstorm workflow includes iterative refinement. When the user selects "Proceed to planning" or indicates they're satisfied with the brainstorm, automatically continue to Phase 4.
Auto-transition trigger: When /workflows:brainstorm completes and the user approves the brainstorm document, proceed immediately to Phase 4 without asking.
Run the planning workflow, which will auto-detect the brainstorm document.
/workflows:plan <feature_description>
The plan workflow will:
docs/plans/Wait for the user to refine the plan. When the user selects "Start /workflows:work" or indicates the plan is ready, automatically continue to Phase 5.
Auto-transition trigger: When the user finishes deepening/reviewing the plan, proceed to Phase 5 without asking.
Execute the plan in the worktree.
/workflows:work <path_to_plan_file>
The work workflow will:
codex-bridge MCP server (delegate_coding_task)Important: The work workflow handles its own PR creation. However, intercept before PR creation to ensure:
After implementation is complete:
# Ensure all changes are committed
git status
# Push the worktree branch
git push -u origin <branch-name>
# Create PR (if not already created by /workflows:work)
gh pr create --title "<type>: <feature title>" --body "$(cat <<'EOF'
## Summary
- <what was built>
- <why it was needed>
## Context
- Brainstorm: docs/brainstorms/<brainstorm-file>.md
- Plan: docs/plans/<plan-file>.md
## Testing
- <tests added/modified>
- <manual testing performed>
---
[](https://github.com/rgardishyan) Generated with [Claude Code](https://claude.com/claude-code)
EOF
)"
Run review and compound workflows in parallel on the PR.
<parallel_tasks>
Code Review:
/workflows:review <PR-number>
This runs multi-agent analysis and creates findings as todos.
Compound Knowledge:
/workflows:compound "Implemented <feature_description>"
This documents solutions and learnings from the implementation.
</parallel_tasks>
After review and compound complete:
git add -A
git commit -m "$(cat <<'EOF'
fix: address code review findings
- <finding 1 addressed>
- <finding 2 addressed>
Co-Authored-By: Claude <noreply@anthropic.com>
EOF
)"
git push
Set PR to automerge:
gh pr merge <PR-number> --auto --squash
Update documentation in parallel:
<parallel_tasks>
a. Dev Log: Create a dev log entry documenting this feature.
bash skill: dev-log
Include: PR link, what changed, key decisions, screenshots if UI work.
b. ADRs: If architectural decisions were made during implementation, record them.
bash skill: using-adr-plugin
Create new ADRs for significant decisions. Link to related existing ADRs.
c. Project docs: Update any project documentation affected by this feature (README, API docs, etc.).
</parallel_tasks>
Clean up worktree after merge:
skill: git-worktree
# Use the skill to remove the worktree
Emit final merge signal to memory (if cases were written during the run):
python3 ${CLAUDE_PLUGIN_ROOT}/skills/memory/scripts/memory.py signal \
<work_case_id> merge 1.0 --source "pr:<PR number>" 2>/dev/null
This closes the feedback loop — the successful merge retroactively upgrades the work phase case's reward, making it more likely to be retrieved in future runs.
Offer memory review (optional):
Ask the user via AskUserQuestion: "Want to run /re:memory-review to inspect the case bank from this run and check for emerging patterns?" Options:
/re:memory-reviewSummary: Present final status to user:
Feature Pipeline Complete!
PR: <PR-URL> (automerge enabled)
Branch: <branch-name>
Brainstorm: docs/brainstorms/<file>.md
Plan: docs/plans/<file>.md
Solutions: docs/solutions/<file>.md (if created)
Dev Log: Updated
ADRs: <new ADRs created, if any>
All documentation updated. Worktree cleaned up.