Skip to content

launch-plan.md

Latest commit

 

History

History
122 lines (83 loc) · 4.52 KB

launch-plan.md

File metadata and controls

122 lines (83 loc) · 4.52 KB

Agent Reliability Kit Launch Plan

Positioning

Agent Reliability Kit helps maintainers verify, harden, and ship AI-agent-assisted codebases before inviting agents, contributors, or CI to touch the repo.

Core promise:

Scan the repository surfaces that most often make AI coding agents fail: instructions, verification commands, README quality, secret hygiene, GitHub Actions safety, MCP/tooling risk, and release readiness.

Audience:

  • maintainers using Codex, Claude Code, Cursor, Gemini CLI, OpenCode, GitHub Copilot coding workflows, or MCP-based local tools;
  • open-source project owners who want contributors and AI agents to follow the same repo rules;
  • engineering teams that need local-first reports before adopting agent-assisted development.

Non-claims:

  • Do not claim the package is already public until the public repo and npm package exist.
  • Do not claim adoption metrics, benchmark wins, security certification, or production usage without evidence.
  • Do not imply the scanner calls agent tools or reads credentials. It scans repository files locally.

Required public links:

  • Public repo: <PUBLIC_REPO_URL>
  • npm package: <NPM_PACKAGE_URL>
  • Documentation: <DOCS_URL>

Launch Goals

  1. Explain the problem clearly: AI agents need reliable repo surfaces, not just better prompts.
  2. Give maintainers a copy-paste quick start that works once the public package is available.
  3. Invite careful feedback before broader distribution.
  4. Keep trust high by being explicit about pre-release status, local-first scanning, and private-data safety.

Launch Readiness Checklist

  • Public repository created and visible at <PUBLIC_REPO_URL>.
  • npm package published or package page available at <NPM_PACKAGE_URL>.
  • Documentation page available at <DOCS_URL>.
  • README has current install instructions and does not contain dead public links.
  • npm run check passes.
  • npm run smoke passes.
  • Release dry run passes without publishing.
  • Demo scan output is reviewed for redaction and private-data safety.
  • Launch copy has no unsupported metrics or "already adopted by" claims.

Recommended Launch Sequence

Phase 1: Soft Announcement

Publish to channels where careful maintainer feedback is likely:

  • personal GitHub release notes or repository discussion;
  • a short X/LinkedIn post;
  • one relevant developer community thread;
  • direct messages to a small number of trusted maintainers.

Message angle:

I built a local-first scanner for the repo surfaces that make AI coding agents succeed or fail.

Success criteria:

  • people understand the category within one minute;
  • at least a few maintainers run it on non-sensitive repositories;
  • feedback identifies confusing findings, missing stacks, or install friction.

Phase 2: Maintainer-Focused Launch

Publish the fuller launch post and demo once the first feedback pass is addressed.

Message angle:

Before letting AI agents work in a repo, check whether the repo is ready for them.

Recommended assets:

  • one screenshot or short clip of the HTML report;
  • a terminal demo using a fixture or public sample repo;
  • a short list of checks: instructions, commands, README, secrets, GitHub Actions, MCP/tooling risk.

Phase 3: Ecosystem Outreach

Share targeted examples for tool-specific audiences without making the product tool-specific.

Examples:

  • "For Codex / Claude Code / Cursor users: are your repo instructions and verification commands agent-readable?"
  • "For MCP users: do your tool configs make permissions and command risk visible?"
  • "For open-source maintainers: can contributors run the same checks before opening a PR?"

Launch Post Outline

  1. Problem: AI coding agents fail when repos lack reliable operating surfaces.
  2. Product: Agent Reliability Kit scans those surfaces locally and writes shareable reports.
  3. What it checks: agent instructions, commands, README, secrets, Actions, AI tooling.
  4. Safety: local-first, no credentials required, redacted evidence.
  5. Quick start: link to <PUBLIC_REPO_URL>, <NPM_PACKAGE_URL>, and <DOCS_URL>.
  6. Ask: run it on a non-sensitive repo and share confusing findings or missing checks.

Copy Guardrails

Use:

  • "pre-release" when public availability is not complete;
  • "local-first";
  • "designed for";
  • "scans repository surfaces";
  • "helps maintainers";
  • "redacts token-like evidence";
  • "agent-neutral."

Avoid:

  • "battle-tested";
  • "trusted by";
  • "secure your AI stack";
  • "prevents all secret leaks";
  • "guarantees safe agents";
  • "available on npm" before <NPM_PACKAGE_URL> is live;
  • "open-source on GitHub" before <PUBLIC_REPO_URL> is live.