Research & Analysis Workflows

Pipeline for grounded research: gather (≥2 sources) → conflict-detect → synthesize-with-provenance. Cites K2 §3.1 (provenance per claim) + K2 §3.2 (multiview retrieval). Reuses the brainstorming (commit ebd136a) as the canonical provenance shape; reuses conflict-detector agent (this plugin) as the contradiction step. 多源研究管道:採集(≥2源)→衝突檢測→帶溯源綜合。 Use when: synthesizing claims from multiple docs/repos/papers, before writing a load-bearing assertion in a spec or design doc, fact-checking a load-bearing PR description, gathering evidence for an architectural decision, validating a claim that contradicts a Phase 0 high-confidence bullet. Skip when: a single authoritative source is sufficient (e.g., 'what is this function's signature' answered by lci:getcontext alone), or the claim is non-load-bearing (typo fix, comment update).

Pi Flow · INIT — create a structured workflow (a DAG of producer/verify nodes coordinating through the filesystem) and stand it up to run as a fleet of efficient pi agents (pi.dev / earendil-works/pi) driven by non-Claude coding-plan models, with Claude Code as the single console. The source of truth is a structured workflow TEMPLATE (.piflow/ /template/); the @piflow/core SDK loads it into a WorkflowSpec and runs it one pi per node. INIT triages your starting point — PORT an existing Claude .js, IMPORT another engine's workflow (n8n/YAML/JSON), or COMPOSE fresh — then builds the template and the per-repo runner. Use to "create/author a pi-flow workflow", "stand up the runner in a repo", "port my Claude workflow to pi", "import an n8n workflow", "run my workflow on a non-Claude model", "pi-runner". To RUN/monitor an existing workflow use piflow-start; to IMPROVE one use piflow-enhance.