Skip to main content

​
Angle

Most prompt libraries let you edit a shared template and hope nothing downstream breaks. STRATT computes the blast radius of every change and posts it to your PR before a human reviewer ever sees the diff.

​
Key Messages

For the design journal (stratt.engineer): @stratt/graph is now live — 35 tests, 8 named failure modes checked on every PR, a full DAG built from import declarations, and a Markdown blast radius report that classifies every affected unit as gate dependency, protected dependency, or standard. Layer 3 of 7. The CI pipeline now enforces what the schema defines. For developer audiences (Show HN / dev community): STRATT treats prompt dependencies like code dependencies. Every prompt unit declares its imports. The graph package builds a DAG, detects cycles (Kahn’s algorithm), resolves all transitive dependencies, and computes which units are affected when you change a shared fragment. Protected agents can’t be silently removed. Gate checkpoints can’t be dropped without a major version bump. The system tells you the blast radius before you merge — not after production breaks. For LinkedIn / professional network: Week 1 of STRATT in public, day 2: shipped the dependency graph layer. 8 failure modes encoded as CI checks. Blast radius computed per PR. Protected agents enforced at merge time. When you treat prompts as versioned protocol units instead of markdown files, you get the same governance guarantees software teams expect from code. The prompt library is now self-governing.

​
Content Opportunities

  1. Design journal entry: “Blast radius before merge” — show the Markdown report format, walk through a real example of changing a shared fragment that impacts 3 chains across 2 domains
  2. Technical post: Kahn’s algorithm for prompt dependency graphs — why topological sort matters when your prompts have explicit import contracts
  3. Thread/short-form: “8 ways your prompt library can fail, and how to catch all of them at CI time” — FM-01 through FM-08 as a numbered thread with one-line descriptions
  4. A2A ops angle: The prompt that built this package referenced a TAD gist, cross-referenced existing types, and specified 8 failure modes — the agent produced 35 passing tests in one session. Link back to the a2a-spec-to-package-prompt-pattern learning.