From devflow
Use during PREVC Execution phase — implements the approved plan using dotcontext agent orchestration (Full Mode) or superpowers SDD/TDD (Lite/Minimal)
npx claudepluginhub nexuz-sys/devflow --plugin devflowThis skill uses the workspace's default tool permissions.
Searches, retrieves, and installs Agent Skills from prompts.chat registry using MCP tools like search_skills and get_skill. Activates for finding skills, browsing catalogs, or extending Claude.
Searches prompts.chat for AI prompt templates by keyword or category, retrieves by ID with variable handling, and improves prompts via AI. Use for discovering or enhancing prompts.
Executes pre-written implementation plans: critically reviews, follows bite-sized steps exactly, runs verifications, tracks progress with checkpoints, uses git worktrees, stops on blockers.
Implements the approved plan task-by-task using dotcontext agent orchestration in Full Mode, or superpowers-driven development in Lite/Minimal Mode.
Announce at start: "I'm using the devflow:prevc-execution skill for the Execution phase."
REQUIRED SUB-SKILL: Invoke devflow:git-strategy
The git strategy skill checks branch protection and creates the appropriate isolation (worktree, branch, or none) based on the project's configured strategy.
plan({ action: "getDetails", planSlug: "<slug>" })
Review the plan from dotcontext. Verify steps are still valid.
Read .context/plans/<slug>.md and review.
Read docs/superpowers/plans/<file>.md and review.
In all modes: raise concerns with user before starting.
If the workflow has autonomy: assisted or autonomy: autonomous:
.context/workflow/stories.yaml existsdevflow:autonomous-loopSkip the rest of Step 3 when autonomous-loop is invoked — the loop handles agent dispatch, TDD, and progress tracking internally.
For autonomy: supervised, continue with the standard execution flow below.
This applies to:
REQUIRED SUB-SKILL: Invoke superpowers:test-driven-development for each task group.
Test types required per task:
| Task touches... | Required test types |
|---|---|
| Pure functions, utilities, business logic | Unit tests |
| API endpoints, DB queries, service boundaries | Unit + Integration tests |
| Auth flows, payments, user registration, onboarding | Unit + Integration + E2E tests |
| UI components (rendering) | Unit + Snapshot tests |
| CLI scripts, hooks | Unit + E2E tests (execute real script) |
When E2E is mandatory:
If the plan does not include test tasks before implementation tasks, stop and fix the plan before executing. </HARD-GATE>
Step 3a — Get agent sequence:
agent({ action: "getSequence", task: "<plan description>" })
Returns ordered sequence. Automatically includes:
test-writer BEFORE any implementation agent (enforces RED before GREEN)code-reviewer (if includeReview is true, which is the default)documentation-writer (at the end)Sequence enforcement: For each task group, test-writer must produce failing tests BEFORE the implementation agent writes code. If the sequence returned by dotcontext places test-writer after an implementation agent, reorder it.
Step 3b — For each step, get agent:
agent({ action: "orchestrate", task: "<step description>" })
Returns recommended agent(s) with docs and playbook path. Uses ONLY task parameter (do not combine with role or phase).
Step 3c — Agent executes step following its playbook
Step 3d — Update progress (continuously, not just at completion):
phaseIdandstepIndexare obtained fromplan({ action: "getDetails" })response.
plan({ action: "updateStep", planSlug: "<slug>", phaseId: "<id>", stepIndex: <n>,
status: "in_progress",
notes: "<decisions made so far, what's next>"
})
On completion:
plan({ action: "updateStep", planSlug: "<slug>", phaseId: "<id>", stepIndex: <n>,
status: "completed",
output: "<step result>",
notes: "<decisions, rationale, test results>"
})
Step 3e — Handoff between agents (when agent changes between steps):
workflow-manage({ action: "handoff", from: "<previous-agent>", to: "<next-agent>",
artifacts: ["src/path/file.ts", "tests/path/test.ts"]
})
Step 3f — After all steps complete:
workflow-advance({ outputs: ["All steps completed", "N tests passing"] })
Returns orchestration + quickStart for the next phase.
Read the relevant agent playbook before each task group:
.context/agents/<agent-name>.md.context/plans/<slug>.mdREQUIRED SUB-SKILL: Invoke superpowers:subagent-driven-development (if subagents available) or superpowers:executing-plans (sequential).
Execute tasks per superpowers workflow without agent guidance.
The Execution phase gate requires:
When gate is met:
workflow-advance() # Moves to V phase
Mark the plan as complete in .context/plans/<slug>.md. Update task checkboxes.
Verify all superpowers review stages passed. Plan document updated with completion status.
For LARGE scale workflows, add checkpoints every 3-5 tasks:
| Thought | Reality |
|---|---|
| "I'll skip the agent sequence" | Agents catch domain-specific issues. Use orchestrate. |
| "Notes in updateStep are optional" | Notes are critical for rehydration after compaction. Always write them. |
| "Agent handoffs are overhead" | Handoffs with artifacts prevent context loss between specialists. |
| "The plan changed, I'll adapt on the fly" | Update the plan document first. Then execute the updated plan. |