FAQ & troubleshooting
Does graft commit to my branch?
No. The merge runs on a temporary beta branch; the stabilized result is copied into your working tree without committing to your base branch. You commit on your own terms. See How sync works.
What if my project has no git repository?
graft uses an internal repo (git_mode = internal) and keeps working. The moment a real git repo is detected on a later sync, graft migrates to tracked. There are no git hooks — this happens only on graft sync.
Can two syncs run at once?
Not on the same workspace. graft holds an exclusive lock per workspace (root, remote, branch). A second sync waits for the first to finish, or resumes it if it stopped on a conflict.
What counts as a "workspace"?
The tuple (root, remote, branch) — not just a directory. One repo can hold several .graft/ sub-trees (separate workspaces), and the same branch in another worktree maps to the same workspace. See Canonical store.
Where does graft store its database?
graft uses one global sqlite database with WAL plus locking for concurrency. The exact path is set by the store implementation (internal/store) and will be documented here once finalized (planned).
Which providers are supported?
Eight active: claude-code, codex, cursor, github-copilot, opencode, roo-code, goose, grok-cli. In addition, antigravity is planned (in the catalog but not yet built into the sync engine, pending a research spike), and gemini-cli is deprecated (previously supported, removed from the active set on 2026-06-15; kept in code and catalog as reference). See Providers. Per-provider pages are added as each lands.
Skills support is available for four active providers: claude-code, opencode, codex, and grok-cli. gemini-cli (.gemini/skills/) is excluded from skill operations because it is deprecated.
Does graft sync skills or slash commands?
graft skill (status/install/sync) is live in v0.0.5: it manages the canonical store at .agents/skills/ and symlinks skills into the four supporting providers (claude-code, opencode, codex, grok-cli). What remains planned: skills as a scope of graft sync (i.e. --scope skills) and slash-command sync. graft is still agents-first for the sync engine.
A sync stopped on a conflict. Did I lose work?
No. The run is saved with its phase and beta branch. Resolve the conflicting path and run graft sync agents --continue. See Resolve conflicts.
Can a teammate resolve a conflict I pushed?
No. Conflicts are per-machine. The resume state (sqlite run row, internal git refs, temporary worktrees) never leaves the machine where the conflict occurred — git push only publishes the committed .graft/ files and provider files at HEAD.
A teammate who pulls your branch cannot run --continue on your behalf; their database has no record of your run. They simply run their own graft sync against the pulled files and resolve any divergence locally.
One important hazard: graft writes git conflict markers into the working canonical files (agent.yaml, instructions.md) when a sync halts — but does not stage or commit them. Running git add . && git commit && git push before resolving them publishes raw markers to teammates. graft now blocks this: graft validate and every graft sync detect unresolved markers in canonical files and produce a blocking error rather than a cryptic YAML parse failure. Resolve or discard the conflicting changes before staging. See Conflicts across machines.