Skip to content

Latest commit

 

History

History
58 lines (46 loc) · 2.91 KB

AGENTS.md

File metadata and controls

58 lines (46 loc) · 2.91 KB

Repository Guidelines

Project Structure & Module Organization

This repository is a documentation-first skill package.

  • eval-advisor/SKILL.md: behavior contract and workflow policies.
  • EVAL_MASTER.md: routing index for all 12 workflows.
  • eval-advisor/references/: conceptual source files (*-guide.md, *-reference.md).
  • eval-advisor/workflows/: execution checklists (*-checklist.md).
  • eval-advisor/templates/: artifact schemas (*-template.md, prompt templates). Keep theory in eval-advisor/references/, procedures in eval-advisor/workflows/, and output formats in eval-advisor/templates/.

Build, Test, and Development Commands

There is no build pipeline or runtime app in this repository. Use lightweight doc validation:

  • rg --files — list tracked files quickly.
  • rg -n "pattern" eval-advisor/ — find and update related guidance.
  • sed -n '1,200p' path/to/file.md — inspect files in small chunks.

Coding Style & Naming Conventions

  • Use concise Markdown with clear headings and short sections.
  • Prefer kebab-case filenames with functional suffixes:
    • *-guide.md, *-reference.md, *-checklist.md, *-template.md.
  • Preserve the no-duplication contract in SKILL.md.

Testing Guidelines

No automated test framework is configured. Treat review as content QA:

  • Verify routing still maps request domain to the correct primary reference.
  • Check workflows stay procedural and templates stay artifact-only.
  • Confirm internal links and file paths resolve.

Commit & Pull Request Guidelines

Use a clear, conventional style:

  • Commit format: imperative, specific subject line (example: Refine error-analysis checklist quantification gates).
  • PRs should include:
    • scope summary,
    • files changed and why,
    • confirmation that no-duplication and reference-first rules were preserved,
    • screenshots only for rendering/layout changes.

Agent-Specific Instructions

Progressive Disclosure Loading Model (Mandatory)

  1. Level 1 (Always load): eval-advisor/SKILL.md and EVAL_MASTER.md.
  2. Level 2 (Only when needed): module-specific execution instructions (eval-advisor/workflows/* and matching workflow section in EVAL_MASTER.md).
  3. Level 3 (Only when required): detailed concept files (eval-advisor/references/*) and artifact files (eval-advisor/templates/*) selected by EVAL_MASTER.md.

Routing and Source-of-Truth Rules

  • First classify domain using EVAL_MASTER.md, then load only that path.
  • Load order: primary reference -> workflow -> template (if needed).
  • Use secondary references only if the selected workflow row lists them.
  • Never treat SKILL.md, workflows/, or templates/ as conceptual source-of-truth when a matching reference exists.

Response Hygiene

  1. Confirm conceptual claims came from the selected primary reference.
  2. Remove duplicated theory that belongs in references/.
  3. Keep workflow/template outputs procedural and concise.