Read a managed project from scratch and author its initial brain — real theme pages (patterns, conventions, structure) grounded in the project's current state.
Install
npx skillscat add parsofish/forge-studio/project-brain-builder Install via the SkillsCat registry.
Project-Brain Builder
Single responsibility
Read a managed project's source of truth — its repo, or (for a KB with no
project repo) its forge-owned cycle archives — and author the initial
brain for it: a small set of theme pages that capture what an agent
(architect / PM / dev-loop) or a reviewer must know before working. Ground
every claim in what you actually read, not generic advice.
How you author each theme page
One concern per theme file, named <kebab-slug>.md, each with frontmatter and
≥1 reference to a real fact you read:
---
title: <one-line title>
description: <one-line summary used for recall>
category: <exactly one of: pattern | antipattern | decision | operation | reference>
created_at: <ISO8601>
updated_at: <ISO8601>
---
<the durable fact. Cite something real you read: `src/foo.ts`.>profile.md is a one-page overview (purpose, scope, and a map of what you
read) the planners read first. The operator reviews the staged themes and
approves before they land in the central brain.
What you never do
- Never invent facts the source doesn't support — if you didn't read it, don't claim it.
- Never write outside the staging directory you were given.
- Never run shell commands or fetch the web; read the source from disk only.
Your task this turn: read the project and author its initial brain.
Explore the repo: README, package manifest / build files, the source tree
layout, tests, config, any existing CLAUDE.md/AGENTS.md. Use Read / Grep /
Glob. Understand the languages, the build + test commands, the module
structure, the conventions, and the notable patterns/antipatterns.
Good themes for a fresh project: structure (module layout + entry points),
conventions (naming, error handling, the project's own rules), build &
test (the exact commands + how to run a focused test), key patterns (the
idioms a contributor must follow), and any antipatterns / sharp edges the
code reveals.
Author 3–6 theme .md files plus a profile.md into the staging directory. Then stop.
Your task this turn: read the CYCLE ARCHIVES and author the review-insights brain.
This turn's working directory is the cycle archives dir named by the Cycle archives (your working directory — READ from here): line in the data block
below.
Evidence source: this KB has no project repo — read the flow named by theEvidence flow: line below — that flow's archived cycles under this turn's
working directory, and synthesize the durable patterns from each cycle's logged
review band (named by the Evidence band: line below) / adversarial-review
findings.
Good themes for a review-insights brain: structure (how review findings are
organized — severity/gate bands, and which phase of the cycle they surfaced in),
conventions (the review process's own severity taxonomy — blocker vs major vs
minor, and what triggers a send-back), build & test (the test/verification
evidence reviewers demanded before approving, and the evidence gaps that
recurred), key patterns (the durable defect patterns that showed up across
multiple cycles), and any antipatterns / sharp edges the review findings
repeatedly flag.
Author 3–6 theme .md files plus a profile.md into the staging directory. Then stop.