Use when the user asks for a structured technical research approach, a spike/proof of concept, feasibility investigation, or research-driven implementation planning. Guides a six-phase workflow (Research, Plan, Iterate Plan, Experiment, Implement, Validate) that generates concrete artifacts: research questions and findings, phased implementation plans with success criteria, experiment reports, implementation logs, and validation reports.
Use when the user asks for a structured technical research approach, a spike/proof of concept, feasibility investigation, or research-driven implementation planning. Guides a six-phase workflow (Research, Plan, Iterate Plan, Experiment, Implement, Validate) that generates concrete artifacts: research questions and findings, phased implementation plans with success criteria, experiment reports, implementation logs, and validation reports.
Validate (/validate) — check each criterion with automated + explicit manual checks.
→ .agents/validation-<slug>.md
Concrete Example
/research auth-system
Produces .agents/research-auth-system.md with content like:
## Findings-`src/auth.py:L42` — JWT decode uses HS256 with a hardcoded secret; risk: key rotation impossible.
-`src/middleware.py:L18` — Token expiry checked only in `auth_required`, not in API routes directly.
## Risks-**HIGH** Secret rotation requires redeploy; tokens cannot be invalidated individually.
## Open Questions1. Does the existing auth middleware support scoped permissions?
2. Which OAuth providers are in scope for initial rollout?
Then /plan auth-system produces .agents/plan-auth-system.md:
## Phase 1: JWT Key Rotation**Success Criteria (Automated):**`pytest tests/test_auth.py::test_rotate_key` passes.
**Success Criteria (Manual):** Old token rejected after rotation without service restart.
## Out of Scope- Social OAuth login (deferred to Phase 2)
Validation Checkpoints and Feedback Loop
After Research: confirm scope is complete and key unknowns are explicit.
After Plan: confirm every phase has success criteria and verification steps.
After Experiment (if used): choose one approach; update plan accordingly.
During Implement: after each phase, run listed automated checks. If fail → fix or update plan with rationale.
Final Validate: report pass/fail per criterion with evidence.
If validation fails: return to Iterate Plan or Implement and re-run validation. Do not continue silently.
Template Assets
Templates are in ${CLAUDE_PLUGIN_ROOT}/skills/research-workflow-management/assets/:
research-template.md — scope, findings, risks, open questions
