AGENTS.md & SKILL.md Operating Architecture

AGENTS.md & SKILL.md Operating Architecture An architecture diagram generated by Archify. Researcher · User Prompt / Task · Architecture component Researcher User Prompt / Task README.md · Project Source of Truth · Antigravity Research Workspace (phdjournal) · Tier 3 · Locked Scope README.md Project Source of Truth Tier 3 · Locked Scope AGENTS.md · Always-On Agent OS · Antigravity Research Workspace (phdjournal) · Tier 2 · L99 Rules AGENTS.md Always-On Agent OS Tier 2 · L99 Rules Session Memory · TODO.md & research-log · Antigravity Research Workspace (phdjournal) Session Memory TODO.md & research-log Skill Registry · YAML Frontmatter Index · Antigravity Research Workspace (phdjournal) · Progressive Disclosure Skill Registry YAML Frontmatter Index Progressive Disclosure Skill Dispatcher · §5.1 Workflow Router · Antigravity Research Workspace (phdjournal) Skill Dispatcher §5.1 Workflow Router SKILL.md Modules · .agents/skills/* (5 Skills) · Antigravity Research Workspace (phdjournal) › On-Demand Skill Sandbox (.agents/skills/) · Tier 1 · On-Demand SKILL.md Modules .agents/skills/* (5 Skills) Tier 1 · On-Demand Subagent Workers · invoke_subagent Roles · Antigravity Research Workspace (phdjournal) › On-Demand Skill Sandbox (.agents/skills/) · Isolated Context Subagent Workers invoke_subagent Roles Isolated Context Manuscript Outputs · journal.pdf & journal-th.pdf · Antigravity Research Workspace (phdjournal) Manuscript Outputs journal.pdf & journal-th.pdf Q1 Quality Gate · DOI · SJR Q1 · 2022+ Rules · Antigravity Research Workspace (phdjournal) Q1 Quality Gate DOI · SJR Q1 · 2022+ Rules Substantive Logs · CONSOLE.md & AI_USE_LOG · Antigravity Research Workspace (phdjournal) Substantive Logs CONSOLE.md & AI_USE_LOG prompt RQs & scope prior state §6 workflow name + desc read SKILL.md spawn role verify claims compile PDF mirror & log Antigravity Research Workspace (phdjournal) On-Demand Skill Sandbox (.agents/skills/) Legend Frontend Backend Database Cloud Security Message bus External

3-Tier Separation

  • • Tier 1 SKILL (.agents/skills/): domain-agnostic methods & prompt codes
  • • Tier 2 AGENTS (AGENTS.md): always-on L99 rules, Q1 gates & §5.1 router
  • • Tier 3 README (README.md): locked project topic, RQ0–RQ5 & 9 constructs

Progressive Disclosure

  • • Startup injects AGENTS.md; Skills expose YAML frontmatter only
  • • Dispatcher loads full SKILL.md strictly when a §6 workflow matches
  • • Complex workflows spawn isolated invoke_subagent worker roles

5 Workspace Skills

  • • academic-prompt-codes & research-background-writer (L99 & Ch.1 CARS)
  • • literature-review-synthesizer & elsevier_format (2022+ matrix & layout)
  • • q1-manuscript-redteam (IMRaD, Reviewer-2 audit & journal targeting)