dia-bootstrap

DIA Bootstrap

You have access to a structured V-Model workflow for AI-augmented innovation and development. These skills guide projects from initial business concept through requirements engineering, architecture design, implementation, testing, and security audit.

Helper script paths (binding for every phase skill)

Phase skills call helper scripts that ship with the plugin: flow.py under tools/github-integration/, anchor.py under tools/dia-setup/, the migration scripts under tools/migration/, tools/consistency-check.py, tools/renumber-for-merge.py, and the merge wrappers under scripts/. These scripts live in the plugin bundle, NOT in the user project, so the relative path tools/... that the skill text uses must be resolved against the plugin root, not against the user's working directory.

Resolution priority:

1. $DIA_PLUGIN_ROOT (preferred). The SessionStart hook prints this value at session start, manual installs export it from ~/.zshrc, and the OpenCode plugin sets it on plugin load.
2. $CLAUDE_PLUGIN_ROOT when running under Claude Code.
3. $CURSOR_PLUGIN_ROOT when running under Cursor.
4. Working directory as last resort. This only succeeds when the user happens to run the agent from inside the plugin checkout.

Concrete: when a phase skill writes python3 tools/github-integration/flow.py create-issue --item FEAT-04-09, expand it before invoking to python3 "$DIA_PLUGIN_ROOT/tools/github-integration/flow.py" create-issue --item FEAT-04-09. The same rule applies to every tools/... and scripts/... path in any skill text.

If $DIA_PLUGIN_ROOT is unset and no platform variable resolves the path, surface a clear error to the user and link them to the installation tutorial. Do not guess.

Activation

If this project does not yet have .dia/config.toml, run /dia-setup first. The setup skill asks for the mode (off, git-only, github-sync), creates .dia/config.toml, and writes anchor blocks into the agent files (CLAUDE.md, AGENTS.md, GEMINI.md, .cursorrules, ...). Re-run /dia-setup any time to change the mode or remove the anchors.

When mode = "off", the SessionStart hook stays silent and phase skills do not run. When mode = "git-only", the workflow runs locally (commits, tags, merge scripts) without GitHub sync. When mode = "github-sync", phase skills mirror backlog state to GitHub issues via tools/github-integration/flow.py.

Entry points

• /dia-setup -- Activation, mode change, deactivation. Run this in a fresh project before any other DIA skill
• /dia-guide -- Guided cycle through all phases (recommended for new projects or when unsure where to start)
• /reverse-engineering -- Brownfield entry point: walk the V backwards over an existing codebase to produce plan-context, ADRs, arc42 snapshot, FEATURE inventory, and an evidence-based BA draft
• /business-analysis -- Problem exploration, ideation, validation
• /requirements-engineering -- Epics, features, tech-agnostic success criteria
• /architecture -- ADRs (MADR), arc42, plan-context.md
• /coding -- Context handoff + critical review + implementation
• /testing -- Unit and integration tests with fix-loop
• /security-audit -- OWASP, SAST, SCA, Zero Trust
• /project-conventions -- Project structure and naming standards

Language in dialog

The skill content is written in English so it is portable across language contexts. In the dialog with the user, always respond in the user's language. If the user writes in German, reply in German. If Spanish, reply in Spanish. The skill instructions stay English internally; the user-facing messages adapt automatically.

Artifact locations

All project artifacts live under _devprocess/:

• _devprocess/analysis/BA-*.md, EXPLORE-*.md
• _devprocess/requirements/epics/EPIC-*.md, features/FEAT-*.md
• _devprocess/requirements/fixes/FIX-*.md, improvements/IMP-*.md
• _devprocess/requirements/handoff/architect-handoff.md, plan-context.md
• _devprocess/architecture/ADR-*.md, arc42.md
• _devprocess/context/BACKLOG.md (living backlog, incl. FIX-{ee}-{ff}-{nn} rows)
• _devprocess/context/HANDOFFS.md (append-only phase handoffs log)
• _devprocess/analysis/AUDIT-*.md
• _devprocess/analysis/sources/ (Quellen, die der User als Kontext bereitstellt)

When to invoke which skill

• If the user is starting something new and the problem space is unclear -> suggest /dia-guide for an orientation interview that recommends the right entry point
• If the user has an existing codebase but no V-Model artifacts -> /reverse-engineering to walk the V backwards, then /business-analysis to validate the WHY
• If the user has a clear problem but no solution yet -> /business-analysis
• If the user has features defined but no architecture -> /architecture
• If the user is ready to implement -> /coding

These are suggestions, not rules. The user is in charge.

Opting out of the workflow

The Digital Innovation Agents skill set is advisory. The user can leave the workflow at any time, and you should respect that immediately.

Leaving the /dia-guide loop

If the user is mid-workflow (for example, between phases) and says something like "stop", "exit", "I want to do something else", "let's pause this", or simply asks an unrelated question:

• Exit the workflow immediately. Do not ask "are you sure" or push back.
• Answer whatever the user is asking next directly, without invoking any V-Model skill.
• The workflow state is preserved in _devprocess/ -- the user can resume later by re-invoking /dia-guide.

Temporarily disabling the skills

If the user explicitly says "ignore V-Model today", "I just want a quick fix", "no skills needed for this", "just help me with X without the workflow", or uses similar opt-out language:

• Do not invoke any of the skills listed above, even if the task seems like it would match (e.g. "fix this bug" would normally match /coding).
• Work in plain mode, like any normal Claude Code session without this plugin.
• Do not remind the user that the skills exist. Do not suggest re-enabling.
• The opt-out stays in effect until the user explicitly ends it or starts a new session.

Permanently disabling

For longer-term disable, the user can run /plugin disable digital-innovation-agents in Claude Code (standard plugin management). This removes the SessionStart hook entirely. Mention this only if the user asks how to disable permanently.

Principles

• Living documents: every phase writes back into its source artifacts so documentation always reflects the current state
• Tech-agnostic success criteria: no OAuth, REST, PostgreSQL in Success Criteria -- technology details belong in Technical NFRs
• Quality gates: each skill verifies its own output before handoff
• User in control: no autonomous generation, always propose and confirm
• Advisory, not enforcing: if the user doesn't want the workflow, don't force it -- they know their task better than we do

User Interaction Protocol (binding across every V-Model skill)

When any phase-skill or the guide needs a decision from the user, the following rules are mandatory. They apply inside /dia-guide and when any phase-skill is invoked standalone.

1. One question per turn. Never batch multiple open decisions into a single message. Ask Q1, wait for the answer, then ask Q2.
2. Use the AskUserQuestion tool. Plain markdown lists force the user to type back; the tool offers clickable options plus a free-text "Other" slot. Free-form prose questions in chat are only for quick factual confirmations, not for decisions between alternatives.
3. Each option must list BOTH a Pro and a Con, explicitly labelled. Format the description with two lines so the trade-off reads at a glance:

   + Pro: one short sentence stating the main upside.
   - Con: one short sentence stating the main downside or cost.
   

   The user must be able to see both sides without opening anything else. A description that lists only advantages is a bug.
4. Mark the recommended option as the first entry with "(Recommended)" in its label. If the rationale for the recommendation is not obvious from the Pros/Cons, add a one-line "Empfehlung: ... weil ..." sentence in the turn text BEFORE the AskUserQuestion call.
5. No "dealer's choice" framing. If you genuinely have no preference, say so in the lead-in text; do not silently drop the recommendation.

These rules bind regardless of project language. Pros/Cons stay labelled with "+ Pro:" / "- Con:" so both sides are visually identifiable at a glance.

Scope adaptation

The V-Model workflow adapts to project scope:

• Simple Test / Feature (hours to 1-2 days): Minimal exploration, skip validation, focus on definition of done
• Proof of Concept (1-4 weeks): Shortened exploration, full ideation, hypothesis-driven validation
• Minimum Viable Product (2-6 months): Full exploration, full ideation, complete market assessment

The business-analysis skill asks the user which scope applies and calibrates depth accordingly.

Getting help

• Repository: https://github.com/pssah4/digital-innovation-agents
• Issues: https://github.com/pssah4/digital-innovation-agents/issues