justsima

A portable agentic research stack for Claude Code: multi-agent deep research (ultradeep), best-on-the-market product research (market-scout), a 5-advisor decision council (llm-council), parallel divergent ideation (adhd), a job-application helper, and a hardened config layer. Evidence-gathering split from judgment, wiki-filed, self-learning.

Multi-agent deep-research pipeline: adaptive parallel explorers + STORM-style questioning + adversarial red-team verification + tiered search (Exa/SearXNG/WebSearch) + optional wiki filing & knowledge-graph. Tunable via program.md.

Find the best X on the market in any category across major retailers. Parallel research explorers + a DETERMINISTIC Python decision-matrix scoring engine + red-team verification + a ranked, segment-aware buying report.

Run a question/plan/decision through 5 advisors that each use a distinct REASONING METHOD (inversion, decomposition, analogy, naive questioning, dependency graphing), then anonymized peer review and a chairman synthesis.

Parallel divergent ideation: spawns N isolated idea branches under different cognitive frames, scores them, prunes traps, and deepens the survivors. For open-ended, high-stakes design/naming/strategy problems.

Generate natural, tailored job-application answers from YOUR own profile (a private, git-ignored profile.md you create from the included template).

The config layer: safety hooks (block dangerous commands, block secret leaks), a session-context hook, a verify-before-stop hook, plus a recommended-settings fragment (permission denylist + workflow flags). No secrets.