chore(agent-setup): centralize shared agent config

.agents/AGENTS.md

@@ -0,0 +1,245 @@
+# Codex Guidelines for Langfuse Python
+
+This is the canonical root agent guide for the repo. The root `AGENTS.md`
+should remain only as a discovery symlink so tools that require that filename
+continue to work while `.agents/` stays the source of truth.
+
+Langfuse Python SDK guidance for fast, safe code changes.
+
+## Maintenance Contract
+
+- `AGENTS.md` is a living document.
+- Update this file in the same PR when repo-level architecture, workflows,
+  verification requirements, release processes, or agent setup conventions
+  materially change.
+- Update this file when user feedback adds a durable repo-level instruction that
+  future agents should follow.
+- Keep root guidance concise, specific, and easy to verify. If repo-wide
+  guidance grows large or becomes task-specific, move that detail into shared
+  skills or future nested `AGENTS.md` files closer to the relevant code.
+- If no durable guidance changed, do not edit AGENTS files.
+
+## Project Overview
+
+This repository contains the Langfuse Python SDK, a client library for
+accessing the Langfuse observability platform. The SDK integrates with
+OpenTelemetry for tracing, provides automatic instrumentation for popular LLM
+frameworks, and exposes a generated API client for the Langfuse platform.
+
+## Project Structure
+
+```text
+langfuse-python/
+├─ langfuse/_client/         # Core SDK implementation built on OpenTelemetry
+├─ langfuse/api/             # Generated Fern API client (do not hand-edit)
+├─ langfuse/_task_manager/   # Background upload and ingestion helpers
+├─ langfuse/langchain/       # LangChain integration
+├─ tests/                    # Test suite
+├─ static/                   # Test fixtures and sample content
+└─ .agents/                  # Canonical shared agent instructions and config
+```
+
+High-signal entry points:
+
+- `langfuse/_client/client.py`: core Langfuse client with OTel integration
+- `langfuse/_client/span.py`: observation/span abstractions
+- `langfuse/_client/observe.py`: decorator-based instrumentation
+- `langfuse/openai.py`: OpenAI instrumentation helpers
+- `langfuse/langchain/CallbackHandler.py`: LangChain integration
+- `langfuse/api/`: generated API surface copied from the main Langfuse repo
+
+## Instruction Design
+
+- Root `AGENTS.md` should cover durable repo-wide expectations only: setup,
+  verification, architecture, security, generated files, and handoff rules.
+- Prefer concrete, testable instructions over vague phrasing. Name the exact
+  command, path, module, or condition whenever possible.
+- Keep stable tone/role guidance separate from task-specific examples. For
+  complex prompts or reusable workflows, place examples in skills or referenced
+  docs instead of bloating the root guide.
+- Add nearby nested guidance only when a subdirectory truly needs different
+  rules. Put the override as close as possible to the specialized code.
+- Use shared skills for recurring task-specific workflows that should not be
+  loaded into context on every task.
+
+## Build, Test, and Development Commands
+
+- Agent environment bootstrap: `bash .agents/scripts/codex/setup.sh`
+- Install dependencies: `bash .agents/scripts/install.sh --all-extras`
+- Sync generated agent shims: `python3 .agents/scripts/sync-agent-shims.py`
+- Verify generated agent shims: `python3 .agents/scripts/sync-agent-shims.py --check`
+- Install pre-commit hooks: `poetry run pre-commit install`
+- Run all tests: `poetry run pytest -s -v --log-cli-level=INFO`
+- Run tests in parallel: `poetry run pytest -s -v --log-cli-level=INFO -n auto`
+- Run one test: `poetry run pytest -s -v --log-cli-level=INFO tests/test_core_sdk.py::test_flush`
+- Format code: `poetry run ruff format .`
+- Lint code: `poetry run ruff check .`
+- Type-check: `poetry run mypy langfuse --no-error-summary`
+- Run pre-commit across the repo: `poetry run pre-commit run --all-files`
+- Build package: `poetry build`
+- Generate docs: `poetry run pdoc -o docs/ --docformat google --logo "https://langfuse.com/langfuse_logo.svg" langfuse`
+
+Minimum verification matrix:
+
+| Change scope | Minimum verification |
+| --- | --- |
+| `langfuse/_client/**` | `poetry run ruff check .` + `poetry run mypy langfuse --no-error-summary` + targeted pytest coverage |
+| `langfuse/api/**` | verify source update path from main repo + `poetry run ruff format .` + targeted API tests |
+| Integration modules (`langfuse/openai.py`, `langfuse/langchain/**`) | targeted tests for the touched integration + lint + latest official provider docs review if behavior or API usage changed |
+| Test-only changes | targeted pytest coverage for the updated tests |
+| Agent setup files (`.agents/**`) | `python3 .agents/scripts/sync-agent-shims.py` + `python3 .agents/scripts/sync-agent-shims.py --check` |
+
+CI notes:
+
+- Linting runs via `astral-sh/ruff-action`.
+- Type checking runs on Python 3.13 with Poetry, `.venv` caching, and the agent
+  shim sync/check step.
+- The main test matrix runs on Python 3.10 through 3.14.
+- Integration CI clones the main `langfuse/langfuse` repo, boots Dockerized
+  services, seeds the server with `pnpm`, and then runs this SDK's pytest suite
+  against that local server.
+- If a change plausibly depends on server behavior, call out whether it was only
+  covered by unit tests locally and whether full CI is the real end-to-end
+  verification path.
+
+## Architecture
+
+### Core Components
+
+- `langfuse/_client/`: main SDK implementation built on OpenTelemetry
+  - `client.py`: core Langfuse client
+  - `span.py`: span, generation, and event classes
+  - `observe.py`: decorator for automatic instrumentation
+  - `datasets.py`: dataset management functionality
+- `langfuse/api/`: auto-generated Fern API client
+- `langfuse/_task_manager/`: background processing for uploads and ingestion
+- `langfuse/openai.py`: OpenAI instrumentation
+- `langfuse/langchain/`: LangChain integration
+
+### Key Design Patterns
+
+- The SDK is built on OpenTelemetry for observability.
+- Spans are the core tracing primitive.
+- Attributes carry trace metadata. See `LangfuseOtelSpanAttributes`.
+- The client batches work and flushes asynchronously to the Langfuse API.
+
+## Generated Files
+
+- `langfuse/api/**` is generated from the main Langfuse repo. Do not edit it by
+  hand unless the task is explicitly about generated client updates.
+- `docs/` output from `pdoc` is generated. Regenerate it instead of editing
+  rendered output directly.
+- Agent/tool shims at `.mcp.json`, `.claude/settings.json`, `.claude/skills/*`,
+  `.cursor/mcp.json`, `.cursor/environment.json`, `.vscode/mcp.json`,
+  `.codex/config.toml`, and `.codex/environments/environment.toml` are local
+  generated artifacts. Update `.agents/config.json` or `.agents/skills/**`
+  instead of editing them by hand.
+- `AGENTS.md` and `CLAUDE.md` at the repo root are compatibility symlinks. Edit
+  `.agents/AGENTS.md`, not the symlink target path directly.
+
+## Configuration
+
+Environment variables are defined in
+`langfuse/_client/environment_variables.py`.
+
+Common ones:
+
+- `LANGFUSE_PUBLIC_KEY` / `LANGFUSE_SECRET_KEY`: API credentials
+- `LANGFUSE_HOST`: API endpoint, defaults to `https://cloud.langfuse.com`
+- `LANGFUSE_DEBUG`: enable debug logging
+- `LANGFUSE_TRACING_ENABLED`: enable or disable tracing
+- `LANGFUSE_SAMPLE_RATE`: sampling rate for traces
+
+Security/config notes:
+
+- Keep credentials and machine-specific secrets in environment variables or
+  local untracked files, never in committed agent config.
+- The shared Claude settings intentionally deny reading `./.env` and
+  `./.env.*`, and they do not auto-approve Bash commands by default. If a task
+  genuinely requires inspecting local env overrides or shell access, get
+  explicit user approval first instead of weakening the default policy.
+- For authenticated MCP servers or provider-specific config additions, prefer
+  secret injection via environment variables rather than committed inline
+  tokens.
+
+## Testing Guidelines
+
+- Keep tests independent and parallel-safe.
+- Do not weaken or delete meaningful assertions just to make tests pass.
+- When fixing a bug, write or update the regression test first when feasible.
+- E2E tests involving external APIs are often skipped in CI. Document when
+  manual coverage is still needed.
+- Use `respx` and `pytest-httpserver` for HTTP mocking when possible.
+- Prefer the narrowest useful test invocation first, then widen coverage when a
+  change touches shared tracing, batching, or provider integrations.
+
+## API Generation
+
+The `langfuse/api/` directory is generated from the Langfuse OpenAPI
+specification via Fern.
+
+Update flow:
+
+1. Generate the Python SDK in the main `langfuse/langfuse` repo.
+2. Copy the generated files from `generated/python` into `langfuse/api/`.
+3. Run `poetry run ruff format .`.
+4. Run targeted verification for any touched endpoints or types.
+
+## Release Guidelines
+
+- Releases are automated via GitHub Actions.
+- The release workflow updates `pyproject.toml` and `langfuse/version.py`,
+  builds the package, publishes to PyPI, and creates a GitHub release.
+- Do not change release/versioning flow without updating this file and
+  `CONTRIBUTING.md`.
+
+## Agent-specific Notes
+
+- `.agents/AGENTS.md` is the canonical root guide.
+- Root `AGENTS.md` is a symlink to `.agents/AGENTS.md`.
+- Root `CLAUDE.md` is a compatibility symlink to `AGENTS.md`.
+- Shared agent/tool config lives in `.agents/config.json`.
+- Shared agent setup documentation lives in `.agents/README.md`.
+- Shared skills live under `.agents/skills/`.
+- `.agents/scripts/` is the home for repo-owned agent bootstrap and sync
+  tooling.
+- `python3 .agents/scripts/sync-agent-shims.py` regenerates tool-specific config
+  shims for Claude, Cursor, VS Code, Codex, and shared MCP discovery files.
+- Tool-specific directories such as `.claude/`, `.cursor/`, `.codex/`, and
+  `.vscode/` remain because those tools discover project settings from fixed
+  paths.
+- Cursor discovery should continue to work through the generated
+  `.cursor/environment.json` and `.cursor/mcp.json` shims plus the root
+  `AGENTS.md` symlink. Do not hand-edit those generated files.
+- This file should stay concise. Anthropic recommends keeping persistent project
+  memory under roughly 200 lines, and both Anthropic and OpenAI guidance favor
+  specific, well-structured instructions over long prose.
+- If future `.cursor/rules/*.mdc` files are added, keep them as thin wrappers
+  around shared `AGENTS.md` guidance or shared skills instead of making them the
+  only source of durable repo guidance.
+- Shared skill index: [`skills/README.md`](skills/README.md)
+- When changing OpenAI or Anthropic integrations, prompts, or documented usage:
+  check the latest official provider docs first, keep prompts simple and direct,
+  preserve clear separation between stable instructions and task-specific
+  examples, and mention any provider-facing verification you did not run.
+
+Official references to start from:
+
+- OpenAI AGENTS guide: <https://developers.openai.com/codex/guides/agents-md>
+- OpenAI prompting guide: <https://developers.openai.com/api/docs/guides/prompting>
+- OpenAI reasoning best practices: <https://developers.openai.com/api/docs/guides/reasoning-best-practices>
+- Anthropic Claude Code memory guide: <https://docs.anthropic.com/en/docs/claude-code/memory>
+- Anthropic Claude Code MCP guide: <https://docs.anthropic.com/en/docs/claude-code/mcp>
+- Anthropic prompting best practices: <https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/claude-prompting-best-practices>
+
+## Git Notes
+
+- Do not use destructive git commands such as `reset --hard` unless explicitly
+  requested.
+- Do not revert unrelated working tree changes.
+- Keep commits focused and atomic.
+
+## Python Code Rules
+
+- Exception messages must not inline f-string literals directly in the `raise`.
+  Assign the string to a variable first if formatting is required.

.agents/README.md

@@ -0,0 +1,129 @@
+# Shared Agent Setup
+
+This directory is the neutral, repo-owned source of truth for agent behavior in
+Langfuse Python.
+
+Use `.agents/` for configuration and guidance that should apply across tools.
+Do not put durable shared guidance only in `.claude/`, `.codex/`, `.cursor/`,
+or `.vscode/`.
+
+## Layout
+
+- `AGENTS.md`: canonical shared root instructions
+- `config.json`: shared bootstrap and MCP configuration used to generate
+  tool-specific shims
+- `scripts/`: shared bootstrap and sync helpers for agent tooling
+- `skills/`: shared, tool-neutral implementation guidance for recurring
+  workflows
+
+## `config.json`
+
+`.agents/config.json` contains five kinds of data:
+
+- `shared`: defaults used across tools
+- `mcpServers`: project MCP servers and how to connect to them
+- `claude`: Claude-specific generated settings inputs
+- `codex`: Codex-specific generated settings inputs
+- `cursor`: Cursor-specific generated settings inputs
+
+Current shape:
+
+```json
+{
+  "shared": {
+    "setupScript": "bash .agents/scripts/codex/setup.sh",
+    "devCommand": "poetry run bash",
+    "devTerminalDescription": "Interactive development shell inside the Poetry environment"
+  },
+  "mcpServers": {
+    "langfuse-docs": {
+      "transport": "http",
+      "url": "https://langfuse.com/api/mcp"
+    }
+  },
+  "claude": {
+    "settings": {}
+  },
+  "codex": {
+    "environment": {
+      "version": 1,
+      "name": "langfuse-python"
+    }
+  },
+  "cursor": {
+    "environment": {
+      "agentCanUpdateSnapshot": false
+    }
+  }
+}
+```
+
+## How Shims Are Generated
+
+`.agents/scripts/sync-agent-shims.py` reads `.agents/config.json` and writes the
+tool discovery files that those products require.
+
+Generated local artifacts:
+
+- `.claude/settings.json`
+- `.claude/skills/*`
+- `.cursor/environment.json`
+- `.cursor/mcp.json`
+- `.vscode/mcp.json`
+- `.mcp.json`
+- `.codex/config.toml`
+- `.codex/environments/environment.toml`
+
+The repo root discovery files remain committed as symlinks:
+
+- `AGENTS.md` -> `.agents/AGENTS.md`
+- `CLAUDE.md` -> `AGENTS.md`
+
+This keeps provider discovery stable while `.agents/` remains the source of
+truth.
+
+## When To Edit `config.json`
+
+Edit `.agents/config.json` when you need to:
+
+- add, remove, or update a shared MCP server
+- change the shared setup/bootstrap command
+- change the default terminal command or terminal label used by generated shims
+- adjust generated Claude, Cursor, or Codex settings that are intentionally
+  modeled in the shared config
+
+Do not edit generated shim files by hand. Edit the canonical files in
+`.agents/` instead.
+
+## Workflow
+
+After editing `.agents/config.json` or `.agents/skills/**`:
+
+1. Run `python3 scripts/agents/sync-agent-shims.py`
+2. Run `python3 .agents/scripts/sync-agent-shims.py --check`
+3. Verify you did not stage generated files under `.claude/skills/` or any of
+   the generated MCP/runtime config paths
+4. Update `.agents/AGENTS.md` or `CONTRIBUTING.md` if the shared workflow
+   materially changed
+
+`bash .agents/scripts/install.sh --all-extras` is the canonical repo install
+flow and wires `poetry install` to `bash .agents/scripts/postinstall.sh`.
+`bash .agents/scripts/codex/setup.sh` uses the same path during agent
+environment bootstrap.
+
+## Adding Shared Skills
+
+Shared skills live under `.agents/skills/`.
+
+Use them for durable, reusable guidance such as:
+
+- repeated SDK maintenance workflows
+- review checklists that should apply across tools
+- repo-specific runbooks that should not live only in a provider-specific folder
+
+Do not use skills for one-off notes or tool runtime configuration.
+
+`python3 .agents/scripts/sync-agent-shims.py` projects shared skills into
+`.claude/skills/` so Claude can discover the same repo-owned skills.
+
+For the skill authoring workflow, see [skills/README.md](skills/README.md).