A personal board of advisors for Claude Code. Six advisors, each built from a structured wiki distilled from that person's own published work, answering from their own method and staying inside their own lane.
Status: v1. One advisor at a time, via /ask-<slug>. No fan-out, synthesiser
or memory yet.
| Path | What it is |
|---|---|
scripts/generate-agents.py |
Reads roster.yaml from the knowledge repo and writes ~/.claude/agents/board-<slug>.md and ~/.claude/skills/ask-<slug>/SKILL.md per advisor. Standalone; run with uv run. |
templates/advisor.md.tmpl |
The advisor agent: mode inference, the mandatory mode and scope lines, the lane rule, citation-following, blocking questions. The wiki's sections 1–5, 8 and 9 are embedded; 6–7 (voice) are read at runtime in counsel mode only. |
templates/ask.md.tmpl |
The /ask-<slug> skill: passes the question to the agent blind, relays the answer verbatim, writes a session file into the knowledge repo. |
templates/berth/ |
Bootstrap for a devcontainer berth, and the checklist for adopting it. |
config.example.toml |
The one config key: where the knowledge repo is. |
docs/ingestion-session-prompt.md |
The pipeline spec: manifest format, raw-file naming, and the nine-section wiki structure every advisor is built to. |
uv run scripts/generate-agents.py --knowledge ~/src/board-knowledgeThen, in a new Claude Code session: /ask-juval [--mode counsel|method] <question>.
Re-run the generator after editing a wiki or the roster.
This repo is generic machinery and knows nothing about any particular user. The advisor wikis, the standing context and the roster live in a separate private repo, and the two are joined at runtime by a path:
--knowledgeflagBOARD_KNOWLEDGEenvironment variable~/.config/board/config.toml./knowledgerelative to the working directory
Agents are build artifacts, generated into ~/.claude/agents/ from the
template here plus the roster there. They are never hand-authored and never
committed.
