Dede

dede

Use this skill to run DogEatDog.DependencyExplorer from an agent. The skill is portable across Codex, Claude Code, and OpenCode because it shells into the repository's CLI instead of assuming a globally installed binary.

Quick Start

1. Identify the dede repository root. Use --repo-root <path> when invoking the wrapper if the current directory is not inside the repository.
2. Use scripts/run-dede.sh for all commands. The wrapper resolves dotnet, finds the repository root, and runs the CLI project directly.
3. Prefer these commands:
   • Scan a workspace: scripts/run-dede.sh --repo-root /path/to/DogEatDog.DependencyExplorer scan /path/to/workspace -o /path/to/graph.json
   • Keep a graph up to date: scripts/run-dede.sh --repo-root /path/to/DogEatDog.DependencyExplorer watch /path/to/workspace -o /path/to/graph.json
   • Serve the explorer: scripts/run-dede.sh --repo-root /path/to/DogEatDog.DependencyExplorer serve /path/to/graph.json --url http://127.0.0.1:5057
   • Export Neo4j CSV: scripts/run-dede.sh --repo-root /path/to/DogEatDog.DependencyExplorer export neo4j /path/to/workspace -o /path/to/neo4j
   • Run impact analysis: scripts/run-dede.sh --repo-root /path/to/DogEatDog.DependencyExplorer query impact --graph /path/to/graph.json --node "<id or name>"
   • Find paths: scripts/run-dede.sh --repo-root /path/to/DogEatDog.DependencyExplorer query paths --graph /path/to/graph.json --from "<node>" --to "<node>"

See references/commands.md for more examples.

Workflow

1. Establish context

• Confirm the workspace root the user wants analyzed.
• Confirm or infer the dede repository root.
• If the repository root is not the current working directory, pass --repo-root.

2. Produce or load a graph

• If the user wants fresh analysis, run scan or export json.
• Prefer watch instead of repeated manual scan calls when the user is actively editing a workspace.
• If the user already has a graph artifact, use it for serve or query.
• Default output path should be explicit, for example ./artifacts/graph.json, rather than relying on implicit files.
• Prefer a root-level .dedeignore file over very long --exclude lists. Use --exclude-file only when the ignore file lives elsewhere.

3. Answer the actual architecture question

• For downstream impact from an endpoint, controller, service, or method, use query impact.
• For an exact connection between two nodes, use query paths.
• For demo mode, run serve and point the user at the local URL.
• Summarize the important counts and findings instead of pasting raw JSON or long command output.

4. Preserve uncertainty

• Do not invent links the scanner did not resolve.
• If the graph contains ambiguous or unresolved edges, say so explicitly.
• When a lookup fails because the graph is missing or stale, regenerate it instead of guessing.

Operational Notes

• The wrapper honors DEDE_REPO_ROOT if set.
• The wrapper also searches upward from the current directory for DogEatDog.DependencyExplorer.sln.
• The skill assumes the repository contains src/DogEatDog.DependencyExplorer.Cli.
• dede automatically loads .dedeignore from the scan root. Agents should rely on that first.
• If dotnet cannot be found, stop and report that the .NET SDK is missing.

Cross-Agent Install Shape

This skill folder is self-contained and can be copied or symlinked into each agent's skill directory.

• Codex: ~/.codex/skills/dede
• Claude Code: ~/.claude/skills/dede
• GitHub Copilot: ~/.copilot/skills/dede or .github/skills/dede
• OpenCode: ~/.config/opencode/skills/dede, .opencode/skills/dede, .claude/skills/dede, or .agents/skills/dede

The agents/openai.yaml file provides OpenAI/Codex interface metadata. The standard SKILL.md folder layout is what Claude Code, OpenCode, and GitHub Copilot consume.

Run scripts/install-all-agents.sh /path/to/DogEatDog.DependencyExplorer to create both repo-local and home-directory links for the major agents.

Source: mitkox/dede — distributed by TomeVault.