| name | research-build |
| description | Use when the user wants a research packet plus a decision-complete implementation plan for a product, feature, algorithm, prompt, bugfix, or refactor, but does not need the full explicit phase-by-phase build loop narration. |
Load the build-loop-auto-research skill when the work is multi-step, integration-heavy, or likely to need iteration.
Use this skill as the shorter packet-oriented path:
- Ground in the repo first.
- Run:
python3 ${CLAUDE_PLUGIN_ROOT}/scripts/research_build.py --task "{{user_request}}" --repo-path "$PWD" --format markdown
- Use the script output as the base scaffold.
- Deepen repo understanding with focused reads across at least 2 relevant files before finalizing any non-trivial plan.
- Add external research only when current external facts materially affect the recommendation, and then restrict it to primary sources, academic institutions, or major research labs.
- Verify integration points and handoffs, especially for APIs, auth, payments, and deployment/runtime-sensitive flows.
- Re-rank confidence as
high, medium, or low.
- Iterate on
medium and low. Spot-check high, then re-calibrate.
Return the final answer in this order:
Bottom line
What I found
Best path
Why this path
Risks / unknowns
Next action