Skip to content

AGENTS.md

Latest commit

 

History

History
151 lines (116 loc) · 5.33 KB

AGENTS.md

File metadata and controls

151 lines (116 loc) · 5.33 KB
Raw

AGENTS.md

Repo-level instructions for coding agents working in DVPE_Daisy-Visual-Programming-Environment/.

This repository is a mixed workspace. It contains the DVPE application, the nested DaisyExamples/ firmware workspace, two Noderr surfaces, legacy agent memory under .agent/, and shared directives/ plus execution/ layers. Work from the smallest relevant subtree and preserve existing connections between these systems unless the task explicitly asks to change the framework itself.

For volatile recency and active-root guidance, read LATEST_PROJECTS.md.

Scope

  • Default assumption: this repo mixes app code, embedded firmware, agent framework files, planning docs, and reference assets.
  • Prefer the smallest possible change set. Do not "normalize" unrelated prompts, docs, or memory files while solving a local task.
  • If work spans more than one root, identify the primary target first and validate only the affected surfaces.

Instruction Priority

When instructions conflict, resolve by descending priority:

  1. Explicit user instruction in the current session
  2. Nearest local AGENTS.md for the target subtree Today this is primarily DaisyExamples/AGENTS.md
  3. This root AGENTS.md
  4. Entry docs and mode selectors such as START_DVPE_SESSION.md, START_FIRMWARE_SESSION.md, .claude/MODE_SELECTOR.md, and the relevant INSTANCE_IDENTITY.md
  5. Tool-specific wrappers such as CODEX.md, CLAUDE.md, CHATGPT.md, GEMINI.md, KILO.md, and OPENCODE.md
  6. LATEST_PROJECTS.md as volatile orientation
  7. docs/plans/ and PLANNING/ as design context, not implementation proof
  8. .agent/daisy_memory/decisions.md and .agent/daisy_memory/patterns.md as strategic memory only

Start Here

  1. Identify the exact target directory before editing anything.
  2. If the target is inside DaisyExamples/, stop and read DaisyExamples/AGENTS.md, DaisyExamples/LATEST_PROJECTS.md, and START_FIRMWARE_SESSION.md before making assumptions.
  3. If the task lives in dvpe_CLD/ or noderr/noderr/, read START_DVPE_SESSION.md, .claude/MODE_SELECTOR.md, and noderr/noderr/INSTANCE_IDENTITY.md.
  4. Read the nearest local docs such as README.md, CHECKPOINT.md, CONTROLS.md, and subtree notes before changing behavior.
  5. If the target is unclear or several roots look plausible, use LATEST_PROJECTS.md to narrow to the active surface.
  6. Use .agent/daisy_memory/decisions.md and .agent/daisy_memory/patterns.md only after local docs, never instead of them.
  7. When the task is about repo automation or agent workflow itself, also read directives/README.md, execution/README.md, and .agent/workflows/.

Repo Map

  • dvpe_CLD/: main React/Vite browser application; no runnable Tauri project
  • noderr/noderr/: primary DVPE Noderr instance
  • DaisyExamples/: nested firmware workspace with its own DaisyExamples/AGENTS.md, LATEST_PROJECTS.md, and noderr/
  • .agent/: legacy memory, skills, workflows, reports, and helper assets
  • .claude/ and _agentic_promts/: mode-switched instruction libraries and prompt assets
  • directives/: repo-level SOP layer describing what to do
  • execution/: deterministic script layer for doing the work
  • docs/, PLANNING/, and project_description/: design and reference context

Pinned Workspace Roots

These roots are first-party work areas and should be considered before assuming the active work lives only in one subtree:

  • dvpe_CLD/
  • noderr/noderr/
  • DaisyExamples/
  • .agent/
  • directives/
  • execution/

Ignore By Default

Do not use these directories as primary evidence unless the task is explicitly about them:

  • .worktrees/
  • .tmp/
  • out/
  • archive/
  • build/
  • dist/
  • __pycache__/
  • nested node_modules/

Generated output in these trees should not override source or docs when judging recency.

Safety Rules

  • Expect a dirty worktree. Preserve unrelated edits.
  • Do not delete or completely rework the existing connections between .agent/, _agentic_promts/, .claude/, noderr/, DaisyExamples/, directives/, and execution/ unless the user explicitly asks for it.
  • Do not treat .agent/daisy_memory/* as the source of truth when local docs or code disagree.
  • Do not treat DaisyExamples/ as generic reference material when you are editing there; it is a nested workspace with its own contract.
  • Prefer thin routing changes over broad prompt rewrites.

Build And Verification

Prefer targeted validation over repo-wide validation.

DVPE app validation

From dvpe_CLD/:

npm test -- --run
npm run build

Run only the smallest relevant command for the surface you changed.

DaisyExamples validation

If the target lives under DaisyExamples/, follow DaisyExamples/AGENTS.md and validate from the specific project directory there.

Framework and repo-contract validation

For root workflow changes such as these files, run the contract test:

node tests/agent-framework-contract.test.mjs

Also run any subtree-specific validation affected by the change.

Documentation Expectations

  • Update LATEST_PROJECTS.md when active work areas or routing guidance change materially.
  • Keep root wrappers thin and route to subtree-local authority when available.
  • Use .agent/daisy_memory/* for continuity, but keep the authoritative instructions close to the code they govern.