-
Notifications
You must be signed in to change notification settings - Fork 0
systems
nam20485
The agent-context template is built from four internal systems that work together to keep AI agents productive and consistent across downstream clones. Each system is a set of files and conventions, not a running service, and each is documented on its own page.
-
Memory and rules system - durable project context and subject-specific conventions under
.agents/ - Validation pipeline - the build, scan, test steps enforced locally and in CI
-
Repository scripts - seven GitHub CLI helper scripts under
scripts/ - App stacks - three pre-defined tech stack profiles referenced by app development plans
The .agents/ directory is the single source of truth for project-specific decisions, conventions, and history. It has three parts: memory.md (durable context with four sections: Current Activity, Completed Work Items, Decisions, Remember To Do), rules/ (ten files, one per subject, covering tools, validation, source control, delegation, practices, coding style, skills, scripts, CI/CD, and AI instruction modules), and skills/ (the gh-issue-tracking-init skill). The discipline is to update as you go, never wait until the end. When content is relocated from AGENTS.md into a rules file, the AGENTS.md section becomes a brief with a link rather than disappearing.
The root validation.ps1 script runs three steps in sequence: build (markdownlint plus relative-link checking), scan (PSScriptAnalyzer for static analysis plus gitleaks for secret scanning), and test (Pester with an 85% coverage gate, JaCoCo XML output, and ReportGenerator HTML report). The GitHub Actions workflow at .github/workflows/ci.yml mirrors this exactly: it installs the same tools, runs validation.ps1, and uploads the HTML coverage report as an artifact. All actions are SHA-pinned. The CI run completes in about 32 seconds.
The scripts/ directory holds seven cross-platform PowerShell 7+ scripts that wrap the GitHub CLI for common operations: authentication bootstrap, label synchronization, PR review-thread management, permission verification, dispatch-issue creation, and remote-instruction-module index regeneration. They share conventions: Set-StrictMode -Version Latest, $ErrorActionPreference = 'Stop', dot-sourcing common-auth.ps1 for auth, -Repo validation against the owner/repo pattern, and a -DryRun switch for previewing actions without mutation.
The .agents/rules/app-stacks/ directory contains three pre-defined tech stack profiles, each named by a slug ID: dotnet-aspire-aspnet-blazor (.NET Aspire web stack), dotnet-avalonia-xplatform-desktop (.NET Avalonia cross-platform desktop), and python-uv-fastapi-vite (Python FastAPI web stack). App development and implementation plans reference these by slug ID to specify the language, tech stack, tools, and packages to use, so the choice of stack is declared once and resolved consistently.
| Path | What it contains |
|---|---|
.agents/memory.md |
Project memory (Current Activity, Completed Work Items, Decisions, Remember To Do) |
.agents/rules/ |
Ten rules files, one per subject |
validation.ps1 |
Root validation script (build, scan, test) |
.github/workflows/ci.yml |
CI pipeline that mirrors validation.ps1
|
scripts/ |
Seven repo-root utility scripts |
.agents/rules/app-stacks/ |
Three tech stack profile files |