obsidian-NotEMD

Imported from Jacobinwwey/obsidian-NotEMD (AGENTS.md). Install upstream with npx skills add Jacobinwwey/obsidian-NotEMD. Copyright stays with the author.

AGENTS.md - Notemd Obsidian Plugin

This is the project-level execution guide for Codex and other repository agents. Treat this file as the authority for build, verification, release, and repository workflow rules. GEMINI.md remains as a background project overview and should not be rewritten to mirror this file.

Project Overview

Notemd is an Obsidian plugin written in TypeScript. It enhances note workflows with LLM-powered processing, concept extraction, Mermaid generation/fixing, translation, web research, and custom one-click workflows.

High-Risk Areas

These files and modules have the highest regression risk and should be treated as coupled surfaces during implementation:

• src/main.ts: plugin entrypoint, commands, settings loading, and task routing.
• src/llmProviders.ts: single source of truth for provider presets, metadata, ordering, and provider-side validation.
• src/llmUtils.ts: provider transport dispatch, request shaping, retry logic, and connection testing.
• src/fileUtils.ts and src/searchUtils.ts: task flows that depend on callLLM() behavior.
• src/ui/NotemdSettingTab.ts: provider settings, workflow builder, and test-connection UI.
• src/workflowButtons.ts and src/ui/NotemdSidebarView.ts: sidebar actions and custom workflow composition.
• src/mermaidProcessor.ts: Mermaid fix and error-detection logic.

When changing one of these areas, check whether a sibling module must also be updated.

Build And Verification

Use these commands from repository root:

npm install
npm run build
npm test -- --runInBand

Verification requirements:

• Do not claim work is complete without fresh build and test output.
• Prefer npm test -- --runInBand for reliable local verification in this repository.
• Run targeted Jest tests first when doing TDD, then rerun the full suite before commit or release.
• Run git diff --check before committing to catch whitespace and patch-format issues.

Pull Request Verification

• .github/workflows/verify-plugin.yml runs build, full Jest, UI-string/render-host audits, lint regression comparison and diff hygiene on Linux and Windows with Node 20. It has read-only permissions and no release or provider credentials.
• Run npm run lint:regressions -- --base-ref origin/main locally. The checker compares individual diagnostics with the merge-base, including renamed and shifted files; a smaller aggregate error count does not excuse new errors.
• Legacy ESLint debt remains tracked. New errors and correctness warnings must pass the ratchet; do not globally autofix unrelated files.
• npm run benchmark:local-kb measures frozen-corpus build/query costs and GC-controlled retained/released heap in a fresh Node process. CI retains the report; timings are observations, not universal performance thresholds. Context token estimates are distinct from provider tokenization.
• Install browser revisions for both playwright and playwright-chromium when preparing a fresh verification environment; the lockfile currently resolves different package versions.
• Workflow presence does not establish required branch protection. Check remote run results before integrating; configure repository protection separately when requested.

Native Consumer Evidence

• scripts/verify-obsidian-host.cjs only operates on a disposable Vault marked with .notemd-host-verification; keep that Vault visible for comparable preview timing.
• scripts/verify-obsidian-persistence.cjs uses the same marker and bundle checks to exercise real rename/replacement races without timing a window. Reload the copied plugin first; it restores injected Vault methods/settings after each run.
• scripts/verify-powerpoint-roundtrip.ps1 saves an editable table change to a separate PPTX, reopens it and renders it with installed PowerPoint. It refuses an existing PowerPoint session.
• scripts/verify-powerpoint-merged-border.ps1 checks the Office fidelity fixture's merged separator through native PowerPoint properties. Run it on exported and reopened PPTX files; full-slide RMSE and DrawingML node presence cannot prove a complete border. It opens read-only, refuses an existing session/report and waits for its COM session to close.
• The Drawnix Plait gate verifies hierarchy and static arrow serialization. It must not be reported as proof of attached cross-branch arrows in the Drawnix application.

Obsidian CLI Checks

When release or integration verification is requested:

obsidian help
obsidian-cli help

Rules:

• Try both commands when the user asks for Obsidian CLI validation.
• If obsidian-cli is not installed or obsidian is blocked by desktop/X11 constraints, report that explicitly.
• Do not claim successful CLI validation unless the command actually executed and produced usable output.

TDD Rules

For behavior changes, bug fixes, or new providers:

1. Write the failing test first.
2. Run the targeted test and confirm it fails for the expected reason.
3. Implement the minimal change.
4. Re-run the targeted test.
5. Re-run the full test suite.

New providers or provider-behavior changes must include tests in the relevant files:

• src/tests/llmProviders.test.ts
• src/tests/llmUtilsProviderSupport.test.ts
• Additional focused tests when adding local validation or UI-facing provider rules

Provider Extension Rules

All new LLM providers must follow these rules:

• Add the preset to src/llmProviders.ts.
• Prefer transport-driven support in src/llmUtils.ts; do not add provider-name branching unless the transport genuinely differs.
• Set provider metadata for:
  • transport
  • category
  • API key mode
  • API test mode
  • default base URL and model
  • setup hint and description
• Update documentation in both README.md and README_zh.md.
• Add or update tests for provider registry coverage, runtime routing, and connection testing.

For OpenAI-compatible endpoints, default to the shared runtime unless the provider requires a dedicated protocol.

If you change the shared OpenAI-compatible runtime:

• Preserve the streaming fallback path for long-running requests on desktop http/https and non-desktop fetch.
• Preserve deep debug coverage for both raw partial bodies and parsed partial stream output.
• Keep legacy exported provider wrappers (callOpenAIApi, callDeepSeekAPI, callMistralApi, callOpenRouterAPI, callXaiApi, callLMStudioApi) delegating to the shared OpenAI-compatible execution path instead of reintroducing buffered-only fallback code.
• Update src/tests/llmUtilsProviderSupport.test.ts to cover both desktop and non-desktop fallback behavior.
• Keep scripts/diagnose-llm-provider.js and scripts/lib/llm-provider-diagnostic.js aligned with runtime protocol behavior so real-endpoint diagnostics match plugin transport semantics.
• Keep the in-plugin settings action Developer provider diagnostic (long request) aligned with the same runtime protocol behavior so UI diagnostics and CLI diagnostics are comparable.

If you change any non-OpenAI-compatible LLM transport runtime (anthropic, google, azure-openai, or ollama):

• Preserve protocol-aware fallback parsing for long-running requests instead of regressing to buffered-only fallback reads.
• Keep partial parsed stream output in shared debug metadata when a fallback attempt is interrupted.
• Update src/tests/llmUtilsProviderSupport.test.ts to cover the affected transport on both success and interrupted-stream paths.

UI, Workflow, And Mermaid Rules

If you touch workflow or sidebar behavior:

• Verify button wiring and workflow chaining tests still pass.
• Preserve the default One-Click Extract workflow unless intentionally changing product behavior.

If you touch Mermaid-related behavior:

• Re-check batch Mermaid fix coverage.
• Re-check workflow paths that can generate Mermaid output.
• Keep Mermaid repair logic scoped to Mermaid code blocks.

Documentation Rules

Documentation is part of the feature. If behavior, provider support, or release workflow changes, update the docs in the same branch.

At minimum, evaluate whether these files need updates:

• README.md
• README_zh.md
• change.md
• AGENTS.md

Release Rules

Before creating a release:

1. Sync version references in repository metadata.
   • package.json
   • manifest.json
   • versions.json
   • any version text in README.md and README_zh.md
2. Run npm run build.
3. Run npm test -- --runInBand.
4. Confirm git status is clean after the release commit.
5. Create/push the Git tag.
6. Create the GitHub Release with release notes.

GitHub Release Notes Format

Every GitHub release body must be fully bilingual:

• Include one complete English section and one complete Chinese section.
• Each language section must be independently readable on its own.
• Do not publish release notes that are only Chinese, only English, or partially mixed with missing mirrored content.
• When using structured release notes, mirror the same core sections in both languages as applicable, such as Highlights, Breaking Changes, New Features, Bug Fixes, Refactors/Chores, and Contributors.

Required GitHub Release Assets

Every GitHub release for this plugin must upload all of the following assets:

• main.js
• manifest.json
• styles.css
• README.md

This is mandatory. Do not publish a release that omits README.md.

Git Safety Rules

• Use non-interactive git commands.
• Never use destructive resets or reverts unless explicitly requested.
• Do not overwrite or discard unrelated user changes.
• If the working tree contains unexpected edits you did not make and cannot explain, stop and ask.
• Prefer feature branches or worktrees for substantial work.

Local Workflow State

• Treat .trellis/ as local workflow state, not as a repo-deliverable source of truth.
• Never delete, overwrite, or bulk-replace .trellis/ during sync, merge, cleanup, release prep, or branch recovery work.
• If a mainline sync step could touch .trellis/, preserve the existing local copy first and keep the repo clean by ignoring it rather than removing it.
• Do not make committed tests, verification scripts, or release gates depend on .trellis/ contents; if a workflow insight needs to affect CI or shipped documentation, mirror that truth into tracked files under the repository proper.

Repository Notes

• main.js is gitignored and is expected to be generated during builds and uploaded as a release asset.
• npm run chronicle:sync-repo-saga and npm run chronicle:update share .cache/repo-saga-* state and must run serially. The script enforces .cache/.repo-saga-execution.lock; remove it only after verifying no repo-saga sync/update run is still active.
• This repository already has broad Jest coverage; treat test regressions as meaningful unless proven otherwise.

Trellis Instructions

These instructions are for AI assistants working in this project.

This project is managed by Trellis. The working knowledge you need lives under .trellis/:

• .trellis/workflow.md — development phases, when to create tasks, skill routing
• .trellis/spec/ — package- and layer-scoped coding guidelines (read before writing code in a given layer)
• .trellis/workspace/ — per-developer journals and session traces
• .trellis/tasks/ — active and archived tasks (PRDs, research, jsonl context)

If a Trellis command is available on your platform (e.g. /trellis:finish-work, /trellis:continue), prefer it over manual steps. Not every platform exposes every command.

If you're using Codex or another agent-capable tool, additional project-scoped helpers may live in:

• .agents/skills/ — reusable Trellis skills
• .codex/agents/ — optional custom subagents

Subagents

• ALWAYS wait for every spawned subagent to reach a terminal status before yielding, acting on partial results, or spawning followups.
  • On Codex, this means calling the wait tool with the subagent's thread id (requires multi_agent_v2). Do NOT infer completion from elapsed time.
  • On Claude Code / OpenCode, this means awaiting the Task/agent tool result before continuing.
• NEVER cancel or re-spawn a subagent that hasn't finished. If a subagent appears stuck, raise the wait timeout (Codex default 30s, max 1h) before judging it broken.
• Spawn subagents automatically when:
  • Parallelizable work (e.g., install + verify, npm test + typecheck, multiple tasks from plan)
  • Long-running or blocking tasks where a worker can run independently
  • Isolation for risky changes or checks

Managed by Trellis. Edits outside this block are preserved; edits inside may be overwritten by a future trellis update.