Create or update the paper folder's 0-lifecycle/5-minimap/5-minimap.tex: the paper IN MINIATURE. Each manuscript paragraph becomes 4-5 sentence-points (one point = one manuscript sentence) tagged with the claim it carries, closed by a lean narrative note and any advisor-feedback line, with the real display floats rendered inline as boxed thumbnails. Venue-shaped (section order + abstract form from _venue/playbook-<venue>), rendered with ref/minimap-template.tex. Use for paragraph minimap, sentence-point spine, evidence anchor, section map, paper-in-miniature, 5-minimap.
Run any paper-lifecycle work. Use `/haipipe-paper enter <paper-path>` or `/haipipe-paper status [paper-path]` to preload an open-needs paper dashboard from STATUS.md, 0-lifecycle, 1-rounds, 0-displays, 0-sections, and git state. Paper lifecycle owns paper-specific story, angle, claims, narrative, displays, minimap, maturity, and dated work rounds; open GAP/NEED items accumulate as probe plans in 1-probe-plans/ and batch-dispatch to /haipipe-probe (the universal evidence gateway for claims; probe calls task/discover during Gather). Direct task/discover verbs available for non-claim utility work. Also parses intent (venue + phase) and dispatches to specialists for writing/revising/rebutting papers. Trigger: paper, enter paper, paper status, open needs, claim gap, figure table gap, round, paper round, work round, write paper, paper pipeline, paper writing, draft paper, revise paper, polish tex, rebuttal, reply to reviewers, probe, probe run, discover, task, evidence, 写论文, 论文流程, /haipipe-paper.
Raw extraction task-folder build specialist. Scaffolds {NN}_<name>/ task-folders under R-series task-groups that extract source tables from Databricks as single parquet files, then process locally with Python. Called by /haipipe-task orchestrator when task-type=raw. Direct invocation works for scoped scaffolding. Cross-references /haipipe-data-raw.
Generate deterministic publication-quality architecture, workflow, and pipeline diagrams from structured JSON (FigureSpec) into editable SVG. Use when user says "架构图", "workflow 图", "pipeline 图", "确定性矢量图", "figure spec", "draw architecture", or needs precise, editable, publication-ready vector diagrams. Preferred over AI illustration for formal architecture/workflow figures.
Generate deterministic publication-quality architecture, workflow, and pipeline diagrams from structured JSON (FigureSpec) into editable SVG. Use when user says "架构图", "workflow 图", "pipeline 图", "确定性矢量图", "figure spec", "draw architecture", or needs precise, editable, publication-ready vector diagrams. Preferred over AI illustration for formal architecture/workflow figures.
Application layer orchestrator (the application umbrella). Routes every session-style workflow in the haipipe stack: intervention lifecycle (seed → pitch → [venue] → claims → narrative → display → minimap → draft → review → deploy → iterate) and research questions (ask). Same stage vocabulary as paper. Venue (SMS, checklist, dashboard, report, etc.) determines which stages fire, claims depth (light/medium/full), and whether gates run. One generic draft skill reads the venue profile — no format-specific sub-skills. Trigger: ask, question, session, intervention, enter, status, seed, pitch, venue, claims, narrative, display, minimap, draft, review, deploy, iterate, round, sms, message, /haipipe-application.
Router and durable lifecycle for the discovery (external-evidence) layer. A discovery is one research topic = one folder, a sibling of a task-folder, running the uniform Plan -> Build(opt) -> Execute -> Report lifecycle across 3 folder types: Search (source = search+read -> sources.md/notes.md), Review (analyze = judge a claim -> verdict.md, or synthesize a field -> landscape.md), Idea (idea -> ideas.md). The 4 capability buckets (search/read/review/idea via arxiv/semantic-scholar/exa/alphaxiv/research-lit/idea-creator/novelty-check) are the Execute-stage workers. Trigger: discover, find paper, lit review, 找idea, 查新, source, verdict, landscape, /haipipe-discovery.
Insight archive orchestrator. Constructs the project's curated permanent knowledge base under examples/<project>/insights/ through review/apply and D/I/K/W card writers. Preferred path: review finished task/probe/discover/narrative/application material, then apply reviewed cards. Low-level writers remain available for explicit data/information/knowledge/wisdom filing. Never executes code, judges probe truth, or triggers probes.