recommend-toolslisted

Recommend and wire **companion tools** for this project — external AI-productivity tools that sit alongside ai-kit's lifecycle skills. ai-kit configures them; it never vendors or auto-installs them. Same trust model as `recommend-rules`: surface, let the user choose, wire only what they pick. ## What a companion tool is A third-party tool *or pattern* that improves the AI coding setup but is **not** part of ai-kit's agile lifecycle. ai-kit owns the *integration glue* (a rules block, a hook, a scaffold), never the tool itself. The current catalog lives at `standards/external/companions.json` — that file is the source of truth for which companions exist, their tiers, detection signals, and conflict checks. The table below summarises; see the JSON for fields, glue paths, and conflict rules. Four are known today: | Companion | Optimizes | Effect | Risk | | --------- | --------- | ------ | ---- | | **graphify** | Input — what the AI reads about the code | Knowledge graph; scoped subgraph per query instead of raw grep. Large token cut on multi-file repos. | Low — invisible infra, just better retrieval. | | **caveman** | Output — how verbosely the AI replies | Compresses responses (drops articles/filler/hedging), technical content intact. ~65% fewer output tokens. | Medium — it is a *communication mode*, opinionated. | | **llm-wiki** | Memory — knowledge from non-code documents | Self-maintaining wiki; ingests specs/transcripts/research into interlinked pages. The wiki, not the r