update

# calcifier:update — detect drift Re-runs the calcify pipeline against the current SKILL.md content (without writing to disk), then unified-diffs the result against the existing `.calcifier/recipe.py`. Surfaces both the SKILL.md sha drift and the recipe textual diff so the user can decide whether to regenerate the recipe. ## When to invoke - The user says "check drift", "did the recipe drift", "update the calcifier recipe", or "is this recipe stale". - After a SKILL.md edit, before running `calcifier exec` again. ## Procedure 1. Confirm the CLI: `calcifier --version`. 2. Locate the SKILL.md and its `.calcifier/recipe.py`. If the recipe is missing, tell the user there is nothing to update and recommend `calcifier:calcifier` instead. 3. Run: ``` calcifier update <abs-path-to-SKILL.md> --compiler compiler --agent-config <agents.json> ``` This operation generates only in a temporary staging directory and reads the on-disk recipe; it does not overwrite the recipe. 4. The CLI prints: - `verdict: STABLE` (no drift) — exit 0. - `verdict: DRIFT` with the SKILL sha mismatch and a unified diff of the recipe body — exit 1. 5. If `DRIFT`, summarize the meaningful changes for the user. The diff is plain Python; common drifts: - New / renamed `@step` functions. - Changed `@step` return annotations (the JSON schema sent to the LLM changes with the type). - Added or removed control flow inside the `@recipe` body. - Changed semantic roles, budgets, parallel fan-out, or subrecipe hashes. 6. Ask the user whether to overwrite. Do not overwrite unless they confirm. The overwrite is `calcifier calcify <skill>` again. ## Output `STABLE` or `DRIFT` verdict plus the unified diff. Exit code reflects drift (0 stable, 1 drift, 2 error).