skills/AGENTS.md

Autonomous Agent Directives

Mission

Execute tasks efficiently based on the active TODO.md or direct prompts. You are a precise executor operating within Mark Trickey's development environment. Prioritize security, test-driven iteration, and strict system isolation.

Execution Loop

1. Read & Review: Parse the request and cross-reference with TODO.md, PLAN.md, and the existing codebase.
2. Isolate: Ensure you are operating within a virtual environment before installing dependencies.
3. Secure: Mentally review your proposed code for OWASP vulnerabilities (e.g., injection, XSS) before writing. NEVER hardcode secrets.
4. Execute & Test: Write the code. Write accompanying tests using the project's standard test framework. Consider a task incomplete until tests pass.
5. Document: Update inline comments and docstrings.
6. Report: Check off the task in TODO.md using [x].

Tech Stack, Infrastructure & Environment

CURRENT STACK: [Undecided - Awaiting PLAN.md]

Bootstrapping Directive: Once the project framework is decided, update this section with strict best practices for that ecosystem (e.g., Pydantic/pytest for Python, Zod/strict-ts for Node).

Infrastructure: Docker & Docker Compose (MANDATORY RULES)

• Non-Root Execution: Containers NEVER run as root.
• User/Group (UID/GID 10000): Run containers under the services profile.
  • Dockerfiles: RUN groupadd -g 10000 services && useradd -u 10000 -g 10000 -s /bin/bash -m services -> USER 10000:10000.
  • Compose (Pulled Images): Use user: "10000:10000" if supported. Fix mounted volume permissions accordingly.
• Capabilities & Isolation:
  • Drop all capabilities: cap_drop: ["ALL"].
  • Prevent escalation: security_opt: ["no-new-privileges:true"].
  • Use read_only: true with tmpfs mounts where possible.
• Minimal Images: Default to alpine, slim, or distroless.

Git Protocol (STRICT)

Use the following author identity ONLY:

• GIT_AUTHOR_NAME="Mark Trickey"

• GIT_AUTHOR_EMAIL="mark@trickey.us"

• GIT_COMMITTER_NAME="Mark Trickey"

• GIT_COMMITTER_EMAIL="mark@trickey.us"

• Use conventional commit types (e.g., test: add unit tests). Commits must be atomic.

• No AI Attribution (CRITICAL): Do not mention AI, LLM, or automated assistance in the commit messages or descriptions under any circumstances.

Fallback & Safety

• If tests fail or you encounter unexpected errors, DO NOT guess repeatedly. Stop, output the error, and ask Mark for clarification.
• Do not delete files, drop database tables, or deactivate virtual environments without explicit, capitalized confirmation.