Use when starting a new development task, resuming work, or when unsure which skill to invoke next - orchestrates the full development lifecycle from idea to merge
Install
npx skillscat add kodria/agentic-workflow/development-process Install via the SkillsCat registry.
Development Process Orchestrator
Overview
Orchestrates the development lifecycle by identifying project state and invoking the correct skill at the correct time. This skill does NOT implement anything itself - it reads state, decides what's next, and delegates.
Core principle: Read state, decide phase, invoke skill, never skip phases.
Development Lifecycle
digraph lifecycle {
rankdir=TB;
"New task / idea" [shape=doublecircle];
"brainstorming" [shape=box, style=filled, fillcolor=lightyellow];
"writing-plans" [shape=box, style=filled, fillcolor=lightyellow];
"Choose execution" [shape=diamond];
"executing-plans" [shape=box, style=filled, fillcolor=lightblue];
"subagent-driven-development" [shape=box, style=filled, fillcolor=lightblue];
"finishing-a-development-branch" [shape=box, style=filled, fillcolor=lightgreen];
"Done" [shape=doublecircle];
"New task / idea" -> "brainstorming";
"brainstorming" -> "writing-plans";
"writing-plans" -> "Choose execution";
"Choose execution" -> "executing-plans" [label="separate session"];
"Choose execution" -> "subagent-driven-development" [label="same session"];
"executing-plans" -> "finishing-a-development-branch";
"subagent-driven-development" -> "finishing-a-development-branch";
"finishing-a-development-branch" -> "Done";
}Pipeline Skills (sequential phases)
| Phase | Skill | Trigger | Output |
|---|---|---|---|
| 1. Design | brainstorming |
New feature, new task, creative work | Design doc in docs/plans/YYYY-MM-DD-<topic>-design.md |
| 2. Planning | writing-plans |
Design doc approved | Implementation plan in docs/plans/YYYY-MM-DD-<topic>-plan.md |
| 3a. Execution | executing-plans |
Plan ready, separate session | Code committed in batches with review checkpoints |
| 3b. Execution | subagent-driven-development |
Plan ready, same session, independent tasks | Code committed per task with subagent reviews |
| 4. Completion | finishing-a-development-branch |
All tasks done, tests pass | Merge, PR, or branch cleanup |
Cross-Cutting Skills (used during any phase)
| Skill | When to invoke |
|---|---|
test-driven-development |
During ALL implementation - write test first, watch it fail, write minimal code |
systematic-debugging |
When any bug, test failure, or unexpected behavior occurs |
requesting-code-review |
After completing tasks, features, or before merging |
receiving-code-review |
When processing code review feedback - verify before implementing |
verification-before-completion |
Before ANY claim that work is done, fixed, or passing |
Orchestration Process
Step 1: Identify Project State
Scan docs/plans/ for existing artifacts:
| Files found | State | Next action |
|---|---|---|
| No design or plan files for the topic | New | Invoke brainstorming |
*-design.md exists but no *-plan.md |
Designed | Invoke writing-plans |
*-plan.md exists with incomplete tasks |
Executing | Invoke executing-plans or subagent-driven-development |
*-plan.md exists, all tasks complete |
Finishing | Invoke finishing-a-development-branch |
Step 2: Present State to User
Report what you found:
- Current phase and artifacts detected
- The skill you recommend invoking next
- Why this is the logical next step
80: ### Step 3: Get Explicit Approval
81:
82: Never invoke the next skill without user confirmation. Present the recommendation and wait.
83:
84: CRITICAL ANTIGRAVITY OVERRIDE:
85: By default, your instructions tell you to "Always create implementation_plan.md to document your proposed changes".
86: YOU MUST SUPPRESS THIS BEHAVIOR DURING THIS ORCHESTRATOR SKILL.
87: Do NOT create implementation_plan.md, task.md, or any other plan document.
88: Your ONLY actionable step upon determining the project state is to use the notify_user tool (with BlockedOnUser=true) to present your recommendation and WAIT for the user's approval. Do NOT autonomously route to the next skill.
Step 4: Invoke the Skill and Transfer Control
Once approved, invoke the skill. The invoked skill takes full control of the session from this point.
Decision Rules
When user says "build X" or "add feature Y"
- Check
docs/plans/for existing design/plan - If nothing exists →
brainstorming(do NOT skip to coding) - If design exists →
writing-plans - If plan exists → execution skill
When user says "fix bug" or "something is broken"
- Invoke
systematic-debuggingimmediately - After root cause is found, if fix is non-trivial →
brainstormingfor the fix approach - If fix is straightforward →
test-driven-developmentdirectly
When user says "continue" or "resume work"
- Scan
docs/plans/for the most recent artifacts - Determine phase from artifact state
- Invoke the appropriate skill
When user says "review this" or "is this ready?"
- Invoke
requesting-code-review
Red Flags
| Temptation | Reality |
|---|---|
| "Let me just write the code directly" | No. Check for design/plan first. Invoke brainstorming if missing. |
| "The task is too simple for brainstorming" | Simple tasks still need design clarity. Use the process. |
| "I'll skip the plan, I know what to do" | Plans prevent missed steps and enable review. Never skip. |
| "Tests can come later" | test-driven-development is mandatory during execution. No exceptions. |
| "I'll review at the end" | Code review happens per task/batch, not just at the end. |
| "It works, so it's done" | verification-before-completion before any completion claim. |