understand-project-structure

name	understand-project-structure
description	Use when starting any LikeC4 modeling task, switching projects, or seeing unknown kind/relationship errors, to re-establish valid element kinds, relationship types, tags, and C1/C2/C3 structure before editing.

Understand Project Structure

Goal

Establish the workspace context before editing so you do not guess kinds, relationships, tags, or the active project.

This skill has two modes:

Mode	What it optimizes for
Preflight for editing	Safe setup before changing elements, relationships, views, or deployment
Onboarding for contributors	Fast understanding of local conventions, sources of truth, and handoffs

When to Use

• Start of any LikeC4 task
• After switching project in a multi-project workspace
• Before C1/C2/C3/view/deployment edits
• When errors mention unknown kinds, relationships, or missing references
• When a contributor is new to this repository or to LikeC4

Source-of-truth order

Always establish context in this order:

1. Active project — list-projects, then read-project-summary for the intended project
2. Semantic taxonomy — projects/shared/SPEC_CHEATSHEET.md and projects/shared/spec-*.c4
3. Project wiring — the project likec4.config.json, model files, and view files
4. Workflow guidance — .github/skills/ tells you how to work, not what the taxonomy is

Important distinction:

• projects/shared/spec-*.c4 = semantic source of truth
• .github/skills/ = workflow guidance
• example projects such as template or spec-showcase = references only, not authoritative taxonomy

Local conventions worth locking in early

Examples actually used here include:

• kinds such as Container_Api, Container_Queue, Container_Database, System_External, Node_Vm, Node_App
• model relationships such as uses, calls, async, reads, writes
• deployment relationship kinds such as http, https, tcp, amqp, ldap, sql, redis, smtp

Do not infer taxonomy from memory or from a nice-looking example file alone.

Preflight for editing

Use this sequence before proposing changes:

1. confirm the active project
2. read the project summary
3. read shared specs and SPEC_CHEATSHEET.md
4. inspect the target project likec4.config.json
5. inspect the nearest model/view files that will actually be edited
6. only then hand off to the next editing skill

MCP checks to re-run after switching projects

• list-projects
• read-project-summary
• search-element or read-element for the target boundary when needed

Re-run them after a project switch because stale context causes the exact failures this skill is meant to prevent: wrong project, wrong kind, wrong parent, wrong FQN.

Onboarding for contributors

When the user is new to the repo, explain the workspace in this order:

1. where the shared taxonomy lives
2. where project-specific models and views live
3. which files are semantic truth vs workflow guidance
4. which next skill should take over once preflight is done

Good onboarding answers explicitly say that:

• .github/skills/ helps with workflow, but does not override shared specs
• projects/shared/spec-*.c4 and the project summary define valid kinds/relationships/tags
• example projects are there to learn patterns, not to redefine semantics

Handoff after preflight

Once context is locked in, hand off deliberately:

• overall C1 → C2 → C3 sequencing → c4-modeling-process
• new element or property → create-element
• new relationship → create-relationship
• static/deployment view → design-view
• dynamic/use-case flow → create-sequence-view
• deployment structure → model-deployment-infrastructure

Short handoff template

Active project and shared taxonomy are validated. The next step is <skill> because the remaining task is <modeling activity>.

If the project summary looks wrong

1. re-run list-projects and verify the intended project ID
2. re-run read-project-summary with the explicit project
3. verify the project likec4.config.json include paths
4. verify shared specs are reachable and match the expected kinds/relationships
5. if still inconsistent, stop and report the context as blocked instead of guessing

If MCP is unavailable

Use an offline preflight instead of stalling:

1. read projects/shared/SPEC_CHEATSHEET.md
2. read projects/shared/spec-*.c4
3. read the target project likec4.config.json
4. read the nearby model/view files that will be edited
5. state what still needs MCP confirmation later (project summary, specific element lookup, relationship graph)

You should still be able to explain local conventions and safe next steps without MCP.

Common mistakes

• ❌ guessing kinds from memory instead of revalidating the active project
• ❌ treating .github/skills/ as the semantic source of truth
• ❌ using example projects as authoritative taxonomy
• ❌ editing C3 without checking the real parent container/system first
• ❌ assuming one read-project-summary run stays valid after switching projects or after major edits

Output

Return a short preflight or onboarding guide that clearly states:

• active project or how to confirm it
• source-of-truth files
• valid local taxonomy examples
• risks avoided by the preflight
• the correct next skill to use