infa-rag

# `infa-rag` — the learned-pattern store Runs `infa2aidp cli rag`. Manages a local store of transformation patterns the LLM-assisted conversion path (`migrate --use-llm`) has produced and, once approved via [`infa-review`](../infa-review/SKILL.md), can reuse on future similar transformations. ## When to use - The user asks "what has the tool learned", "show me the pattern store", "how many approved patterns do we have". - Moving approved patterns between environments (export on one, import on another). - Resetting the store, e.g. before a clean benchmark run. ## Actions ```bash PYTHONPATH="${INFA_ENGINE:-$HOME/.aidp-infa-migrator/engine}" python3 -m infa2aidp.cli rag stats PYTHONPATH="${INFA_ENGINE:-$HOME/.aidp-infa-migrator/engine}" python3 -m infa2aidp.cli rag list --approved-only PYTHONPATH="${INFA_ENGINE:-$HOME/.aidp-infa-migrator/engine}" python3 -m infa2aidp.cli rag export -o rag_export.json PYTHONPATH="${INFA_ENGINE:-$HOME/.aidp-infa-migrator/engine}" python3 -m infa2aidp.cli rag import -i rag_export.json PYTHONPATH="${INFA_ENGINE:-$HOME/.aidp-infa-migrator/engine}" python3 -m infa2aidp.cli rag clear ``` | Action | Flags used | Output | |---|---|---| | `stats` | — | `Entries: N Approved: M Retrievals: K` | | `list` | `--approved-only` (optional), `-i`/`-o` unused | Up to 20 entries: `[id] <transformation_type> -- approved/pending -- used Nx` | | `export` | `-o` (default `rag_export.json`) | `Exported to <path>` | | `import` | `-i` (required for this action) | `Imported from <path>; total entries now N` | | `clear` | — | `Cleared <N> entries` — **irreversible**, confirm with the user first | ## What "approved" means here An entry becomes approved through the [`infa-review`](../infa-review/SKILL.md) flow, not directly through `rag`. This command only reads/writes/manages the store; it does not itself judge quality. ## When to use this vs. just re-running migrate - Only relevant to the `--use-llm` path — the rule-based converters never touch this store. - `clear` before a benchmark so results aren't skewed by prior-run pattern reuse; leave it alone in normal operation so accuracy improves run over run. ## After this No downstream step — this is bookkeeping, consulted implicitly by future `migrate --use-llm` runs.