ai-engineering-fluency: Instructions file for GitHub Copilot

pnpm Publishing via OIDC (Trusted Publishing)

This repository publishes the CLI package to npm using GitHub Actions OIDC trusted publishing — no NPM_TOKEN secret is required. The CLI uses pnpm as the package manager; publishing is done via pnpm publish.

Required setup

1. Job permission: id-token: write must be set on the publish job.
2. setup-node: Use registry-url: 'https://registry.npmjs.org' so pnpm is configured for the registry.
3. Publish command: Always include --provenance so the OIDC token is exchanged automatically:

- name: Publish to npm
  run: pnpm publish --provenance --access public

Common mistake to avoid

Do not set NODE_AUTH_TOKEN to an empty string:

# ❌ WRONG — explicitly clears the token and causes ENEEDAUTH
run: NODE_AUTH_TOKEN="" pnpm publish

# ❌ ALSO WRONG — requires a secret that isn't needed
run: pnpm publish
env:
  NODE_AUTH_TOKEN: ${{ secrets.NPM_TOKEN }}

# ✅ CORRECT — OIDC token is exchanged automatically via --provenance
run: pnpm publish --provenance --access public

Setting NODE_AUTH_TOKEN="" overrides the auth that setup-node configures and causes ENEEDAUTH failures even though id-token: write is set.

---

Workflow Security: Validating Untrusted User Input

Overview

Workflows in this repository that are triggered by untrusted user input (issue bodies, PR descriptions, comments, branch names, etc.) must validate that input for hidden characters and potential prompt injection before processing it.

This is especially important for workflows that pass user content to AI/LLM systems (e.g. GitHub Copilot agents), but also applies to any automated processing where a malicious actor could influence the workflow's behavior.

The Central Validation Script

.github/workflows/validate-input.sh is the single, authoritative script for this check. It detects:

Threat	Description
Bidirectional Unicode control characters	Trojan Source attack (CVE-2021-42574) — makes text look different to humans vs. AI
Zero-width / invisible characters	Hidden text injected between visible characters, invisible to human reviewers
Unicode tag characters (U+E0000–E007F)	Completely invisible; can encode arbitrary ASCII instructions
Unicode variation selectors	Can steganographically encode hidden data
HTML comment blocks (<!-- ... -->)	Stripped by GitHub's renderer but fully visible to LLMs processing raw Markdown
Non-printable control characters	Unexpected control bytes that may confuse parsers