Skip to content

Latest commit

 

History

History
240 lines (142 loc) · 9.53 KB

File metadata and controls

240 lines (142 loc) · 9.53 KB

Initialization and Invocation

This guide is for Cursor, Copilot, and Claude Code invocation: slash commands, #prompt: fallbacks, and side-by-side examples per assistant. It does not duplicate install steps, session rhythm, or the canonical prompt library.

Need Open
Install, upgrade, verify Installing into your project
Copy-paste prompts Session prompt standard
Script sequences and checklists Daily runbook
Cursor / Copilot / Claude Code command syntax This page

How to Run Assistant Commands

Lines like /sdlc-spdd-init are not shell commands. Do not paste them into a terminal. They are prompts you send in the AI chat inside your editor (Cursor Chat/Agent, GitHub Copilot Chat, or Claude Code).

Cursor

  1. Open the target application folder in Cursor (not the orchestrator repo, unless you are developing the framework itself).
  2. Open Chat or Agent (Ctrl+L / Cmd+L, or the chat panel).
  3. Type / and pick sdlc-spdd-init from the list, or type /sdlc-spdd-init and press Enter.

The command runs as a chat message. The agent reads .cursor/commands/sdlc-spdd-init.md and follows that skill.

GitHub Copilot Chat

  1. Open the target application in VS Code (or another Copilot-enabled editor).
  2. Open Copilot Chat.
  3. Type /sdlc-spdd-init and send.

If slash commands do not appear:

#prompt:sdlc-spdd-init

Or: Command PaletteChat: Run Prompt → choose sdlc-spdd-init.

Claude Code

  1. Open the target application folder in Claude Code.
  2. Type / and pick sdlc-spdd-init from the list, or type /sdlc-spdd-init and press Enter.

The command runs as a chat message. Claude Code reads .claude/commands/sdlc-spdd-init.md and follows that skill, with CLAUDE.md loaded automatically as project memory.

Shell commands vs assistant commands

Kind Example Where you run it
Shell (terminal) ./scripts/sdlc-spdd/sdlc.sh next, ./scripts/sdlc-spdd/sdlc.sh claim <WORK-ID> Terminal in the target project
Assistant (chat) /sdlc-spdd-whereami, /sdlc-spdd-init, /sdlc-spdd-plan @requirements/foo.md Cursor, Copilot, or Claude Code chat input

All /sdlc-spdd-* lines in the docs are assistant commands unless they start with ./ or cd.

First-Time Application Setup

Default install (from this orchestrator repo):

./scripts/setup-agent-prompts.sh --target /path/to/your/project --all
./scripts/verify-project-install.sh --target /path/to/your/project

Full install options, upgrade path, and what gets created: Installing into your project.

Initialize Context Inside the Application

After the scaffold files are installed, open the target application in your AI coding tool and invoke initialization.

Cursor:

/sdlc-spdd-init

GitHub Copilot Chat:

/sdlc-spdd-init

If slash commands are not visible in Copilot Chat, use one of these options:

#prompt:sdlc-spdd-init

or open the Command Palette and run:

Chat: Run Prompt

Then choose sdlc-spdd-init.

Claude Code:

/sdlc-spdd-init

Expected result:

  • Stack markers are detected.
  • Memory and harness files are created or preserved.
  • Existing application code is not changed.
  • The assistant recommends the next SDLC-SPDD skill.

Start or Resume an Agent Session

Script sequence and checklists: Morning or Start-of-Session Check in Daily runbook. Resume prompt wording: Start of session in Session prompt standard.

Before asking a new agent to continue previous work:

  1. Orient:

    ./scripts/sdlc-spdd/sdlc.sh next
    

    In chat: /sdlc-spdd-whereami

  2. Claim or resume and open a session brief:

    cd /path/to/your/project
    ./scripts/sdlc-spdd/sdlc.sh claim FEAT-001-order-status-api
    ./scripts/sdlc-spdd/sdlc.sh resume FEAT-001-order-status-api --phase code
    ./scripts/sdlc-spdd/sdlc.sh start
    
  3. Paste the Resume Prompt from agent-context/sessions/current-session.md. Do not paraphrase — the brief embeds Resolved Context (phase files, extensions, Work ID artifacts, area-filtered index rows) and the Resume Prompt points at only those files. See Session prompt standard.

Optional canvas sync before step 2:

   ./scripts/sdlc-spdd/resync-agent-session.sh --target . --work-id FEAT-001-order-status-api --check-only

How to Start Work

Canonical prompt wording: Session prompt standard and Triage. Below: Cursor, Copilot, and Claude Code invocation examples for common entry points (plain language, files, Jira, GitHub, bugs). The slash-command syntax (/sdlc-spdd-* @file) is identical across all three assistants.

Use one Work ID for each unit of work. Good IDs include:

  • FEAT-001-order-status-api
  • BUG-003-null-order-total
  • REF-002-split-billing-service
  • DOC-004-runbook-refresh

Start from a plain-language idea

Cursor:

/sdlc-spdd-plan Create an endpoint that returns the current status for an order by ID.

Copilot Chat:

/sdlc-spdd-plan Create an endpoint that returns the current status for an order by ID.

Claude Code:

/sdlc-spdd-plan Create an endpoint that returns the current status for an order by ID.

Start from a requirement file

Create a requirement document first:

requirements/order-status-api.md

Then invoke planning:

/sdlc-spdd-plan @requirements/order-status-api.md

Start from a Jira issue

Use the Jira key or URL and include the project context that the assistant needs.

/sdlc-spdd-plan Jira ABC-123: add order status lookup. Link this work to https://jira.example.com/browse/ABC-123 and create the canvas as FEAT-001-order-status-api.

If the issue text is not available to the assistant, paste the acceptance criteria:

/sdlc-spdd-plan Jira ABC-123

Acceptance criteria:
- GET /orders/{id}/status returns the latest status.
- Unknown orders return 404.
- The response includes orderId, status, and updatedAt.

Start from a GitHub issue

/sdlc-spdd-plan GitHub issue https://github.com/example/orders-api/issues/42. Create a REASONS Canvas and link the issue in Metadata.

Start from a bug report

/sdlc-spdd-plan BUG: checkout fails when an order has no discount. Use BUG-003-null-discount-checkout as the Work ID. Identify the likely code path, tests, and safeguards before coding.

Start from an existing canvas

If a canvas already exists, continue from it instead of starting over.

/sdlc-spdd-architect @spdd/canvas/FEAT-001-order-status-api.md

or:

/sdlc-spdd-code @spdd/canvas/FEAT-001-order-status-api.md operation T01

How to Ask Questions That Keep Context

Always include the Work ID and point to the active artifacts. More examples and anti-patterns: During session and Anti-patterns in Session prompt standard.

Good (Cursor, Copilot, or Claude Code):

For FEAT-001, read @spdd/canvas/FEAT-001-order-status-api.md and @agent-context/features/FEAT-001-order-status-api/progress-log.md. What should I do next?

Avoid:

What now? / Continue. / Can you fix it?

Invoking the SDLC-SPDD Skills

The same /sdlc-spdd-* syntax works in Cursor, Copilot, and Claude Code.

Skill Invocation (Cursor / Copilot / Claude Code) Use when
Initialize /sdlc-spdd-init First time in a target application
Plan /sdlc-spdd-plan @requirements/file.md Convert requirement, Jira issue, or GitHub issue into a canvas
Architect /sdlc-spdd-architect @spdd/canvas/WORK-ID.md Harden the canvas before coding
Code /sdlc-spdd-code @spdd/canvas/WORK-ID.md operation T01 Implement one approved operation
Review /sdlc-spdd-review @spdd/canvas/WORK-ID.md Review changes against the canvas
Prompt update /sdlc-spdd-prompt-update @spdd/canvas/WORK-ID.md Update the canvas first when requirements, acceptance criteria, or behavior intent change
Retro /sdlc-spdd-retro @spdd/canvas/WORK-ID.md Capture reusable learnings
Sync /sdlc-spdd-sync @spdd/canvas/WORK-ID.md Reconcile implementation reality with the canvas

Daily Invocation Pattern

Full step order and part ownership: Workflow. Typical command sequence:

/sdlc-spdd-plan @requirements/<topic>.md
/sdlc-spdd-architect @spdd/canvas/<WORK-ID>.md
/sdlc-spdd-code @spdd/canvas/<WORK-ID>.md operation T01
/sdlc-spdd-review @spdd/canvas/<WORK-ID>.md
/sdlc-spdd-code @spdd/canvas/<WORK-ID>.md operation T02
/sdlc-spdd-review @spdd/canvas/<WORK-ID>.md
/sdlc-spdd-retro @spdd/canvas/<WORK-ID>.md
/sdlc-spdd-sync @spdd/canvas/<WORK-ID>.md

Use review after each coding operation when the change is risky, touches shared behavior, or changes user-visible behavior.

If the requirement or intended behavior changes midstream, update the canvas before coding:

/sdlc-spdd-prompt-update @spdd/canvas/<WORK-ID>.md

If a refactor changes only internal structure, review the change and then sync the canvas:

/sdlc-spdd-review @spdd/canvas/<WORK-ID>.md
/sdlc-spdd-sync @spdd/canvas/<WORK-ID>.md

End each meaningful session by persisting memory:

./scripts/sdlc-spdd/sdlc.sh capture --summary "<summary>" --validation "<tests>" --next "<next command>"