Agentic Builder Lab

Agentic Builder Lab

One build session, five artifacts. This skill takes the receipts from a real build and produces the content stack that ships to frankx.ai.

The flywheel

A single build session produces:

MDX build-log entry — long-form record at content/builds/[slug].mdx (rendered by /agentic-builder-lab/[slug] on the frankx.ai site).
LinkedIn draft — short post that points back to the build log.
Demo brief — 60-90s video script with shot list.
Mermaid diagram — the architecture or agent flow, ready to embed.
Reusable prompt — one artifact from the build added to the prompt pack so the next builder can copy it.

The point is leverage. One session, five published surfaces. No artifact is decorative — each one points to the next.

Voice override

This skill depends on frankx-brand, but overrides its default studio voice for every output it produces. The override is loaded from resources/voice-spec.md after frankx-brand is loaded, so the override wins.

The override rules in short:

Technical authority. Results first, claims never.
No spiritual, consciousness, or transformation language.
No exclamation marks. No hype adjectives.
Lead with what was built, the tool, and the outcome.
Show failures honestly — humility is on-brand.

Banned terms (lint at exit): soul, awaken, transformation, alignment, journey, intentional, sacred, energy, vibration, frequency, studio (as metaphor), elevate (as metaphor), unlock (as metaphor).

Encouraged terms: architect, agent, verification, orchestration, stack, primitive, harness, scaffold, ship, repo, PR, eval.

Full spec: see resources/voice-spec.md.

How the skill runs

Step 1 — Load voice override

Read resources/voice-spec.md.
Treat its rules as binding for every artifact this skill writes.
If frankx-brand was loaded earlier in the session, its studio voice is now suppressed for outputs of this skill.

Step 2 — Collect inputs

Required fields:

build_name — short label (becomes the title).
status — one of live, shipping, wip, paused, archived.
stack — list of tools and what each one does in the build.
key_learning — the one thing the builder would tell another builder.

Optional fields:

demo_url — live demo URL.
repo_url — public repo.
day_number — if this is part of a build-in-public counter.
outcome_metric — one quantifiable result (latency, cost, users, PRs merged).
slug — kebab-case slug. If not provided, derive from build_name.

If any required field is missing, ask the user for it in one batched message. Do not invent values. Do not proceed until all four required fields are filled.

Step 3 — Generate the five artifacts

Load each template from resources/ and fill it in:

Output	Template	Destination
MDX build log	`resources/mdx-template.md`	`content/builds/[slug].mdx`
LinkedIn draft	`resources/linkedin-template.md`	`drafts/linkedin/[slug].md`
Demo brief	`resources/demo-brief-template.md`	`drafts/demos/[slug]-brief.md`
Mermaid diagram	`resources/diagram-prompt.md`	`drafts/diagrams/[slug].mmd`
Prompt pack entry	`resources/prompt-pack-conventions.md`	`skills/projects/agentic-builder-lab/resources/prompt-pack/[slug].md`

The MDX file is the canonical artifact. The other four reference it.

Step 4 — Lint

After all files are written, grep each one for banned terms. If any banned term appears, rewrite that artifact before exit. Suggested check:

# studio, elevate, unlock are banned as metaphors only. Literal usage
# ("recording studio", "elevate floor 3", "unlock the door") is fine —
# review each hit against resources/voice-spec.md before rewriting.
BANNED="soul|awaken|transformation|alignment|journey|intentional|sacred|energy|vibration|frequency|studio|elevate|unlock"
grep -InE "\\b($BANNED)\\b" \
  content/builds/[slug].mdx \
  drafts/linkedin/[slug].md \
  drafts/demos/[slug]-brief.md \
  drafts/diagrams/[slug].mmd \
  skills/projects/agentic-builder-lab/resources/prompt-pack/[slug].md

Lint must pass on a clean exit.

Step 5 — Report

Print the five file paths and a one-line summary of each. Do not push, do not commit. The caller decides what to ship.

Example invocation

User: /build-log

> build_name: Antigravity Codex Bridge
> status: shipping
> stack:
>   - Antigravity: agent harness
>   - Codex CLI: parallel verification
>   - Claude Code: PR review
>   - Vercel: deploy
> key_learning: Two agents disagreeing on the same PR catches more bugs than one agent running twice.
> day_number: 14
> outcome_metric: 3 PRs shipped in one afternoon
> demo_url: https://example.com/demo
> repo_url: https://github.com/frankxai/example

Skill produces:

content/builds/antigravity-codex-bridge.mdx
drafts/linkedin/antigravity-codex-bridge.md
drafts/demos/antigravity-codex-bridge-brief.md
drafts/diagrams/antigravity-codex-bridge.mmd
skills/projects/agentic-builder-lab/resources/prompt-pack/antigravity-codex-bridge.md

Files in this skill

agentic-builder-lab/
├── SKILL.md                                 ← this file
└── resources/
    ├── voice-spec.md                        ← voice override (load first)
    ├── mdx-template.md                      ← build-log MDX scaffold
    ├── linkedin-template.md                 ← 9 LinkedIn post combos
    ├── demo-brief-template.md               ← 60-90s video brief
    ├── diagram-prompt.md                    ← 3 Mermaid diagram templates
    ├── prompt-pack-conventions.md           ← how to add a prompt
    └── prompt-pack/
        └── example-agents-md.md             ← starter prompt: AGENTS.md template

Notes for the model running this skill

Read voice-spec.md before generating any artifact. Re-check it before the lint step.
If the user already has a build log at content/builds/[slug].mdx, ask whether to update it or write a new entry with a different slug.
The prompt pack is cumulative — never overwrite an existing entry without asking.
The MDX frontmatter must validate against the schema in mdx-template.md. If any field is missing, fall back to null, not an empty string.