Method
Method
Implementation-Planning-Driven AI-Assisted Programming. AI coding is fast but drifts and loses context. This AI-native engineering method constrains AI so every step is verifiable and deliverable. The discipline itself is executed by AI, as a chain of skills.
Observational Convergence (RC)
The engineering discipline and the product skills both grow from Observational Convergence (RC). Its core move is observational locking: grow a causal chain node by node, verifying each node before trusting it. The three red lines and the fidelity mechanism are therefore principled, not arbitrary.
Engineering chain (14)
An auditable spine, plus two entries, two reverse-engineering on-ramps, and two supports. Describe what you want; the chain runs unattended.
Entries: engineer-job (full build), engineer-next (resume). Reverse on-ramps: engineer-cloner, engineer-legacy-recon. Supports: engineer-coach, engineer-advisor.
Blueprint trail
Every step leaves a persisted artifact. The blueprint is the single source of truth. On context bloat, open a fresh session and point it at the blueprint instead of pushing through.
- REQUIREMENTS.md Bounded contexts, business events, dependencies, state machines
- CONTEXT.md System overview, data models, API contracts, milestone dependency tree
- FRONTEND-DESIGN.md Page tree, component tree, state management, UI state machines, design tokens
- POC-MANIFEST.md Mock-to-real evolution map (optional)
- .agents/ job.state.json / progress.json / qa-latest.md, cross-session progress persistence
Three red lines
Non-negotiable. Each with the cost of breaking it.
No work without a blueprint
No code before requirements, architecture, and frontend design produce their docs. Otherwise drift starts at step one.
No consolidation without verification
No commit or consolidation before inspection. Otherwise drift compounds.
Rebuild on chaos
When a session tangles, reset context and rebuild from the persisted blueprint. Otherwise context rots and state hallucinates.
Fidelity mechanism
Why trust what AI produces. Every recon and POC files an honest ledger, graded three ways. No fake backend source, no raw asset copying, no mock passed off as real.
- CLONE-FIDELITY.md engineer-cloner / observable-exact / inferred / unobservable
- RECON-FIDELITY.md engineer-legacy-recon / stated / inferred / gap
- POC-FIDELITY.md engineer-poc / real interaction / mock data / placeholder
Test gate
engineer-qa is the single source of truth for the test gate, auto-triggered after a feature. Unit, integration, E2E. Branch coverage on changed code stays at least 90 percent, with a global no-regression ratchet. Non-UI projects degrade to black-box acceptance. The verdict lands in .agents/qa-latest.md: pass, needs fix, or rebuild.
22 skills
Every installable piece of the method. Engineering (14) is the spine; product and RC are the same-root ecosystem.
Same-root ecosystem
Repo and install
The method is 22 open-source, installable skills. Replicable, made concrete as installable.
git clone https://github.com/iannil/skills && cd skills
./install.shOr npx iannil/skills install all. Standard installer: npx skills add iannil/skills --skill '*'.