What problem does this solve?
AgentDiff setup is currently mostly global. That is convenient for personal use, but teams may want repo-scoped configuration that can be reviewed, versioned, and enforced per project.
The main question: should agentdiff configure support project-level agent setup, or should project-level guidance live in files like AGENTS.md, .cursor/skills/, and repo-local config only?
What would you like to see?
Explore and add project-level configuration support.
Possible CLI shape:
agentdiff configure --scope global
agentdiff configure --scope project
agentdiff configure --scope project --agents cursor,codex
Possible project-level outputs:
- repo-local AgentDiff config under
.agentdiff/ or .git/agentdiff/
- repo-local agent guidance such as
AGENTS.md
- project Cursor skill installation
- repo-local VS Code workspace settings where applicable
- clear docs for which integrations cannot be made project-local because the agent only supports global hooks
Open questions
- Which agents actually support project-local hooks/settings?
- Should
AGENTS.md become the canonical cross-agent instruction surface?
- Should project-level configure write files into the worktree, or only into
.git/agentdiff/?
- How do we avoid surprising users by modifying repo files?
Acceptance criteria
- There is a documented distinction between global and project-level setup.
- The CLI can run project-level setup without overwriting unrelated files.
- Project-level setup is safe to review in a PR when it writes worktree files.
- Agents that cannot be configured per-project are reported clearly.
Alternatives considered
Only keep global configure and add docs. That is simpler, but weaker for teams and experiments where every agent should receive the same repo-level instruction.
What problem does this solve?
AgentDiff setup is currently mostly global. That is convenient for personal use, but teams may want repo-scoped configuration that can be reviewed, versioned, and enforced per project.
The main question: should
agentdiff configuresupport project-level agent setup, or should project-level guidance live in files likeAGENTS.md,.cursor/skills/, and repo-local config only?What would you like to see?
Explore and add project-level configuration support.
Possible CLI shape:
Possible project-level outputs:
.agentdiff/or.git/agentdiff/AGENTS.mdOpen questions
AGENTS.mdbecome the canonical cross-agent instruction surface?.git/agentdiff/?Acceptance criteria
Alternatives considered
Only keep global configure and add docs. That is simpler, but weaker for teams and experiments where every agent should receive the same repo-level instruction.