Bcp
Tempest — a cross-harness software factory: opinionated build & QA workflows as portable Agent Skills for Claude Code and pi.
npx -y skills add mtthsnc/tempest --skill bcpAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its author says it does
Copied from the file, not written here
Use when brand or marketing work needs to be on-brand and traceable — set up a Brand Context Protocol (BCP) for a business, capture brand truth, or produce/score a deliverable (landing page, deck, email, ad copy) against the brand. Triggers on "set up a brand", "make this on-brand", "scaffold a BCP", "brand context protocol", "produce on-brand <thing>", "is this on-brand". Bridges the dev stack to the BCP template at github.com/mtthsnc/bcp.
SKILL.md
3.6 KB, 812 tokens by cl100k_base, as published. Nobody here has run it
bcp — bring a Brand Context Protocol into the work
Overview
The bridge from this software factory to BCP (Brand Context Protocol) — a git-native standard
where a brand is written down as cited, dated truth that humans and AI agents build from, so
output is on-brand by construction and carries a trace. BCP lives in its own repo
(github.com/mtthsnc/bcp); this skill scaffolds it into a project and drives its skills. It does not
reimplement BCP — the brand repo is the source of truth, owned by you, not locked in tooling.
This is a STARTER skill — a generic spine. Rewrite the scaffold source and the per-deliverable flow to match how your business runs.
Procedure
1. Find or create the brand
- Look for an existing BCP in the project: a
harness/brand/tree (optionally withharness/config.json). - If none exists and brand work is needed, scaffold one: fetch the template into a brand repo,
e.g.
git clone --depth 1 https://github.com/mtthsnc/bcp <brand-dir>(ornpx degit mtthsnc/bcp <brand-dir>), setbrand.nameinharness/config.json, and run its./install.shto wire the BCP skills for pi and Claude Code. Work from the brand repo root so itsAGENTS.md/CLAUDE.mdand skills load.
2. Capture or extend Brand Truth
- Drive BCP's
brand-truthskill: turn raw input (interviews, audit, existing assets) into cited, dated truths inharness/brand/— positioning (01-imprint), voice + refusals (02-voice), design tokens (03-design), lenses, architecture. Append-only; supersede, never overwrite.
3. Produce a deliverable
- Use a BCP contract skill (e.g.
landing-page) to generate the asset intoharness/output/<slug>/with atrace.mdciting the truth ids it used. Draw only from Brand Truth — if a fact is missing, add it viabrand-truthfirst rather than inventing it.
4. Check before shipping
- Run BCP's
brand-checkskill to score the output against the brand (cited violations, written back to the trace, logged to09-loops), and run./scripts/brand-check.shfor the structural gate. - Recurring conflicts become proposed new truths/refusals + a dated decision — human-approved.
Common mistakes
- Brand work with no BCP. Producing copy/design without a source of truth yields off-brand, untraceable output. Scaffold or locate the brand repo first.
- Inventing brand facts here. This skill orchestrates; it does not author truths. Missing facts
go into Brand Truth via
brand-truth, with a source. - Shipping without a trace or a check. Every deliverable carries a trace and passes
brand-check. - Editing truths in place. BCP is append-only — supersede, don't overwrite.
Guardrails
- The BCP repo is the source of truth; do not duplicate or override brand facts inside this stack.
- Every brand deliverable must carry a trace and pass
brand-checkbefore it's called done. - Keep the brand in its own git repo — owned and portable, not rented inside someone's tool.
- Defer brand authoring, scoring, and production to BCP's own skills (
brand-truth,landing-page,brand-check); this skill only routes to them.