infa-review

# `infa-review` — the human gate Runs `infa2aidp cli review`. Two actions: `generate` (produce a review file from an export) and `import` (record human decisions back into a report). This is the mechanism that keeps low-confidence conversions from shipping unnoticed. ## When to use - Immediately after [`infa-migrate-mapping`](../infa-migrate-mapping/SKILL.md). - The user asks "what needs a human look", "show me the low-confidence items", "what got flagged for review". **Do NOT treat a clean `infa-migrate-mapping` exit code as "nothing to review".** The migrate command's own exit code reflects whether it *ran* without crashing, not whether every mapping converted with confidence. `infa-review generate` re-parses the same input and re-runs the conversion pipeline to find every item scored LOW, MEDIUM, or MANUAL — this is the step that actually reveals that count. ## Step 1: generate the review file ```bash PYTHONPATH="${INFA_ENGINE:-$HOME/.aidp-infa-migrator/engine}" python3 -m infa2aidp.cli review generate \ -i <path-to-export> \ -o ./review/review_items.json ``` Console: `Review file: <path> (<N> item(s) to review)`. `N` is every transformation-level conversion scored LOW, MEDIUM, or MANUAL confidence — this is a superset of the `# REVIEW REQUIRED` markers baked into generated notebooks, since it also catches lower-confidence items that still produced runnable (if uncertain) code. ## Step 2: a human works the review file The review file is JSON. Each item needs a `decision`: `approved`, `edited` (with the corrected content), or `rejected`. Walk the user through it item by item rather than bulk-approving — that defeats the point of the gate. Cross-reference anything involving an unsupported construct (taskflow/Mapping Task/parameter-set/worklet reference, or a source/target tier the generator can't resolve — see [`infa-migrator-overview`](../infa-migrator-overview/SKILL.md)) against the known-gap list before asking the user to decide; some of these can't be fixed by editing the generated cell, only by handling the orchestration piece manually outside this tool. ## Step 3: import the reviewed file ```bash PYTHONPATH="${INFA_ENGINE:-$HOME/.aidp-infa-migrator/engine}" python3 -m infa2aidp.cli review import \ -i ./review/review_items.json \ -o ./review ``` Console: `Approved: X Edited: Y Rejected: Z`. Writes `review_report.md` under the output directory. ## Flags | Flag | Notes | |---|---| | `action` | positional, `generate` or `import` | | `-i, --input` | for `generate`: the export path; for `import`: the reviewed review-file path | | `-o, --output` | for `generate`: the review-file path to write; for `import`: a directory (`review_report.md` is written inside it), default `./review` | ## What "review" does not cover - It does not re-run [`infa-migrate-mapping`](../infa-migrate-mapping/SKILL.md) or regenerate the notebook from an `edited` decision automatically — an edit recorded here needs to be applied to the actual `.ipynb` by hand or by re-running migrate with adjustments (`--custom-rules`, `--params`). - It has no concept of taskflow/Mapping Task/parameter-set/worklet review, because the parser produces nothing for those constructs to review in the first place — they simply don't appear. ## After this - Every `rejected` item is a mapping the user has decided not to trust as generated — track it separately, it will not reconcile. - Once approved, proceed to [`infa-reconcile`](../infa-reconcile/SKILL.md) to prove correctness against real data, then [`infa-deploy`](../infa-deploy/SKILL.md).