content-production

Source-of-truth for TimelinesAI long-form content articles + the test harness that produces the numbers, screenshots, and tool traces those articles cite.

The hard rule: every claim in a published article has a corresponding file under evidence/ that was produced by running a script under tests/. If the evidence doesn't exist, the claim doesn't ship.

The articles themselves are deployed from InitechSoftware/timelinesai-landing (typically as React components under src/components/guides/<slug>/). This repo is the upstream draft, the test harness, and the audit trail — not the production code.

Layout

articles/<slug>/
README.md — article-specific running notes
content/ — drafts (markdown), final copy mirror (en.ts)
prompts/ — recipe-book prompts as standalone .md files
tests/ — runnable validation scripts
evidence/ — committed test outputs (anonymized)

Currently one article: articles/whatsapp-mcp-guide/ — source for timelines.ai/whatsapp-mcp-guide.

Setup

cp .env.example .env
# Fill in TLA_TOKEN_* values from 1Password or the team's secrets store.# .env is gitignored.

Tokens are TimelinesAI Public API bearer tokens. Get them from app.timelines.ai → Integrations → Public API → Copy (or Generate).

Test runs

cd articles/whatsapp-mcp-guide
python tests/<script>.py
# Outputs land in evidence/<scope>/<timestamp>.json

All test scripts:

• Read tokens from ../../.env (repo root).
• Write raw output to evidence/<scope>/<filename>.raw.json (gitignored — may contain PII).
• Write anonymized output to evidence/<scope>/<filename>.json (committed). Anonymization replaces real phone numbers with +1 555 0100..0299, real names with Alice, Bob, Carol..., real message UIDs with MSG-001.., etc.
• Log timestamps in ISO format.

Anonymization rules

Per feedback_no-real-pii-in-public-examples: real phone numbers, message UIDs, customer names, and email addresses NEVER ship in evidence/ committed files. The raw .raw.json files capture the truth for audit; only the scrubbed .json versions are committed.