CommunityWriting & Editinggithub.com

FICC Research Report Writing Skill

Uses a given FICC or macro topic and a corpus (research library, data, old drafts) to produce an analysis outline, material index, and supporting decision-making. Not for short monthly pieces, Q&A, data lookups, or rewriting finished articles.

What is FICC Research Report Writing Skill?

This skill, `ficc-hand`, acts as a central router for composing structured, evidence-based FICC (Fixed Income, Currency, and Commodities) and macro research reports. It orchestrates an eight-step workflow: framing the topic, inventorying the provided corpus, digesting materials, mapping relationships, developing thesis and counter-views, building a hierarchical outline and a materials index, performing acceptance checks, and finally handing off to the `ficc-voice` agent for prose writing. The main output is two files: a view outline in a specific format (`<report-name>-提纲.md`) and a materials index (`<report-name>-资料索引.md`) containing pointers to the source data. The skill explicitly does not write article body text, perform editorial reviews, or handle short pieces, Q&A, or data lookups—these remain the domain of `ficc-voice` or other agents. It is designed for advanced agentic workflows (e.g., Claude Code, Cursor, Codex) and relies on specialized sub-agents (`ficc-mapper`, `ficc-outliner`) and scripts (`scripts/tieout.py`) to ensure cross-referencing accuracy and traceability of all claims, numbers, and institutions back to the source materials.

Works with✓Claude Code~Codex CLI~Cursor
npx skills add https://github.com/changmingcn/FICCWriting/tree/HEAD/skills/ficc-hand

Installed? Explore more Writing & Editing skills: steipete/notion, langchain-ai/langchain, bytedance/podcast-generation · View all 6 →

Ask in your favorite AI

Open a new chat with this agent skill pre-loaded.

Documentation

ficc-hand

This file is T0's router. T0 reads FICCHand.md, and that suffices; agents read only their own card and the templates the packet names (files under references/), never FICCHand.md. Only the outliner reads FICCOutline.md in full, to build the tree and to check (every agent reading the rule files in full was the largest avoidable cost on the bill).

Does: from topic and corpus, run eight steps (FICCHand.md 3): framing, corpus inventory, digests, map, thesis and counter-views, outline and materials index, acceptance, handoff. Deliver <report-name>-提纲.md (view tree, format per ~/.claude/skills/ficc-voice/FICCOutline.md) and <report-name>-资料索引.md (pointers, not excerpts; format in references/materials-manifest.md). The outline-and-index step has two halves (FICCHand.md 2, 5a and 5b): ficc-mapper builds the materials index, ficc-outliner builds the tree; only ficc-outliner builds trees. After handoff, ficc-voice takes over per references/handoff.md: tree building, placement and outline maintenance to ficc-outliner (FICCOutline), relations to ficc-structurer (FICCStructure), style, sentence patterns and tone to FICCVoice.

Does not: write body text, run body-text acceptance (tone check, cold read), deliver articles, or sweep-edit; all are in ficc-voice. Adds no gates: research routes and the four gates belong to research-agent, corpus retrieval to ficc-research-kit, task records to first-secretary, dispatch and caps to dynamic-workflow. Each rule lives in its own home; FICCHand.md only points there.

Proof: the self-test samples must report unmatched items, and scripts/tieout.py --self-test must report PASS (per group the actual unmatched count, at least one, as expected). Each run's acceptance keeps three results (outline check, tieout, counter-view). Recovery: install with install.sh in the task directory, undo with rollback.sh. If the user changes a view after handoff, change the tree in ficc-voice first; do not rerun this skill. Priority: the user's current request and CLAUDE.md outrank this skill; where research-agent or ficc-voice rules differ, the rule's home wins.

Agents: ficc-mapper (T1E, this skill's card, ~/.claude/agents/ficc-mapper.md) fills the map (脉络图), builds the materials index (资料索引) and revises it per index findings from acceptance. ficc-outliner builds the tree (reads FICCOutline.md in full plus references/outline-from-research.md), checks the outline in check mode at acceptance, and fixes outline findings in rule or maintain mode. Other seats use existing cards (digester, collator, dissenter, refuter, scout). Script: scripts/tieout.py checks that every number, date and institution in outline claims and 〔标题〕 lines is in the materials index or the ledger.

Individual skills in this repo

This repo contains 1 individual skill — each has its own dedicated page.

Related Skills