SKILL PI LOCAL

Dev Workflow

Multi-pass AI workflow for shipping high-quality code. Use when the user is about to finalize a change, wants a standards review, or asks to ship/workflow/review their code. Covers plan review, self-review, standards pass, documentation, and PR creation.

Overview

Multi-pass AI workflow for shipping high-quality code. Use when the user is about to finalize a change, wants a standards review, or asks to ship/workflow/review their code. Covers plan review, self-review, standards pass, documentation, and PR creation.

This is a Pi-local skill shipped inside the Dev Workflow package. Install the parent Pi package, reload Pi, and the skill becomes available alongside the package's extension surface.

Parent Surface

Parent docs: Dev Workflow

This skill is documented together with the parent Pi package at /pi/dev-workflow, including package-level commands, extension files, and install guidance.

Workflow Overview

text 
1. Plan Review      → /workflow:plan       → Challenge the plan before code gains momentum
2. Park Reviewer    → [manual]           → Save implementation details, hold reviewer in reserve
3. Self-Review      → /workflow:self       → Implementor rereads own code with fresh eyes
4. Standards Pass   → /workflow:standards  → Lint, types, conventions, pre-commit hygiene
5. CI Check         → /workflow:ci         → Fetch CI status, analyze failures, distinguish ours vs flakes
6. Verify & Loop    → [manual]           → Verify locally, wake reviewer, iterate 2-6 if needed
7. Documentation    → /workflow:docs       → Explain the why and how for future readers
8. Ship             → /workflow:ship       → Verify CI green, discover PR context, atomic commit, open PR

Step Details

Before writing code, challenge the plan itself. Have the AI reread the plan and surrounding code with fresh eyes, looking for bugs, ambiguity, and conflicts. Update the plan based on findings.

Command: /workflow:plan (append extra context like /workflow:plan focus on auth module)

When AI finishes implementation and outputs "details of what it did":

This step is manual orchestration. No command for it.

Force the implementor (same AI session that wrote the code) to reread all new and modified code with fresh eyes. Catch obvious bugs before the independent reviewer spends time on them.

Command: /workflow:self

Run the coding standards and pre-commit cleanup. This is the policy and hygiene pass:

Command: /workflow:standards

Before manual verification, check CI for the current branch using the separate ci-status pi package when installed. If it is unavailable, use get_ci_status and ci_fetch_job_logs only if the current harness exposes those tools; otherwise ask the user to install ci-status before proceeding.

Primary commands:

Orchestration command:

The ci-status extension auto-watches CI on startup and after git pushes. Failure notifications appear automatically. Covers GitHub Actions and CircleCI (set CIRCLECI_TOKEN for CircleCI enrichment).

Command: /workflow:ci (orchestrated) or /ci-detail (direct interactive UI)

The engineer verifies everything locally (backend, frontend, etc.). Then wakes the waiting reviewer session. If the reviewer finds issues, paste findings back into the implementor and repeat steps 2-5 until satisfied.

This step is manual orchestration. No command for it.

Make the outcome legible. Document all updated code, especially new additions, in simple, visual, first-principles-driven language. Explain why changes were made.

Command: /workflow:docs

Finalize and ship the work:

Never compromise by excluding files that are part of the change. Everything touched must be improved.

Command: /workflow:ship

  • No local imports (check circular imports)
  • No unnecessary getattr(), use hasattr() only if needed
  • No overly large try/except blocks
  • Structured logging in optimo_ apps
  • No hardcoded strings/numbers where structured fields should be used
  • Use TypedDict instead of loose dict with Any
  • Ruff must be happy with all files
  • No string-based type hints
  • Never use typing.cast() — it's a code smell
  • Don't repeat fixtures in tests
  • Use Django ORM reverse relations to avoid unnecessary model imports
  • Be pedantic about type hints, avoid Any
  • Use ast-grep where helpful
  • /ci — quick status overview in the widget area
  • /ci-detail — interactive TUI view grouped by CI provider and workflow/cycle, Tab and cycle switching, native pickers, in-place refresh, automatic failure focus, detail view, and log access
  • /ci-logs — pull failure logs for a specific job
  • /workflow:ci — guides the AI through the full CI check: run /ci or /ci-detail, analyze each failure (ours vs flake), propose fixes, summarize.

Principles

Placeholders (repo-specific)

Resources

No extra references or helper scripts are documented for this skill.

Installation

Install the parent Pi package, then reload Pi so the skill is registered locally.

bash 
# From the agent-skills-marketplace repo root
pi install "$PWD/pi-packages/dev-workflow"

# From the Diversio monolith root
pi install "$PWD/agent-skills-marketplace/pi-packages/dev-workflow"

/reload

After install, see Dev Workflow for the parent package's commands and extension behavior.