chemaclass/agnostic-ai/sync

sync

Emits one set of specs to every configured target in its native format.

Steps

1. Run agnostic-ai sync. It writes only the files whose content changed, so git status after a no-op run is empty.
2. Read the summary. Each line names a target and the paths it wrote.
3. Show the user the diff: git diff --stat (or git status --short when generated files are gitignored).

Checking instead of writing

agnostic-ai sync --check compares planned output against what is on disk and writes nothing. Exit 0 means no drift; non-zero lists the stale files with counts. Add --diff to print the change per file. That is the CI gate:

- run: agnostic-ai sync --check --diff

The CLI must be installed on the runner first. The CI page (https://agnostic-ai.org/docs/ci/) shows the npm dev dependency and the pinned scripts/install.sh setups.

When --check fails, decide by what changed:

• A spec was edited and the generated file lags (listed as out of date): run agnostic-ai sync and commit both.
• The CLI was upgraded and output shifted: run sync, review the diff, and commit it as a regeneration with no semantic change.
• Someone hand-edited a generated file (listed as edited locally): their edit is about to be lost. Move the change into the matching spec under .agnostic-ai/, then sync.

Notes

• agnostic-ai sync --watch re-emits on every save, touching only affected targets. Useful while authoring several specs.
• agnostic-ai sync --backup keeps a .bak beside each overwritten file; agnostic-ai revert restores them. Without --backup, revert deletes generated files.
• agnostic-ai status reports spec counts, configured targets, the last sync time, and whether output is stale. agnostic-ai doctor is the stricter pre-merge check.
• A warning about an unsupported kind means the target has no surface for it (for example hooks on a target with no hook system). Silence it per project with on-unsupported: silent.
• Never resolve drift by editing generated files. The spec is the source of truth.