create-brief

create-brief

Interactive skill that produces a research brief at {output_dir}/brief.md. The brief is the single source of truth that the deep-research skill reads to dispatch parallel research agents.

A research project has this layout:

{output_dir}/
├── brief.md              # this skill's only output
├── research/             # one file per topic, written by /deep-research
│   └── {topic-slug}.md
└── synthesis/            # syntheses across the corpus
    ├── general.md        # the whole-corpus synthesis, rewritten each round
    └── {axis-slug}.md    # per-axis cuts, written later by /synthesize

Why this matters

The deep-research workflow runs in rounds: pick a domain, plan one parallel agent per topic, each producing an exhaustive markdown report (1500–4000 words, dense external citations) by doing fresh web research, then synthesise across all topics into a tight executive summary. The first round is typically 5–15 topics; later rounds add coverage based on what the synthesis surfaces. Rounds are append-only — earlier rounds are never rewritten.

The brief is the difference between research that produces tight, comparable topic files and a sprawl. An agent reading a single topic line in the brief should be able to start working immediately, with no follow-up questions. That bar is what this skill exists to enforce.

Workflow

Step 1 — Resolve output root folder

Before opening the wizard, determine the root folder where research output lives in this repo. Run the cascade below and stop at the first match. The cascade resolves the root only — the per-project subfolder is {root}/{domain-slug}/, confirmed in Step 2 Q7.

1. Consumer-defined skill. If a skill named sagan-load-output-folder is registered in the user's environment, read its SKILL.md body for a line of the form output_folder: <path> and use that path. The plugin ships a template at sagan-load-output-folder/SKILL.template.md that consumers copy into their own repo (or .claude/skills/, or any plugin) and edit.
2. Harness instructions. If the harness instructions in your context document an output folder for AI output or research output (any naming the local convention uses — e.g. ai-output-folder, research-output, docs-research-folder), use that path. Read the instructions and infer from context; don't insist on a specific marker or filename.
3. Repo heuristic. Scan the repo for evidence of an existing research folder: directories containing brief.md files, or common patterns like meta/research/, docs/research/, research/, ai/research/. If you find a clear single candidate, use it. If multiple plausible candidates exist, list them and ask the user to pick.
4. Default. Fall back to meta/research/.