module named `nudge_builtins` (not `builtins`) because the stdlib `builtins` is loaded before user code and would permanently shadow a local `builtins.py`; handlers/matchers referenced by allowlisted string name only
each exists, and the
measurements behind them. It is loaded automatically by Claude Code.
**Contributors: start with `CONTRIBUTING.md`** — the short version (setup, boundaries, house rules,
testing habits). This file
description" --body "Closes # "
```
4. **Before merging**, ensure:
- CI pipeline passes (all checks green)
- Code has been reviewed if required
- No merge conflicts with `main`
**IMPORTANT:** Never push directly to `main
that take required parameters or handle workspace state. |
| [NAMING_CONVENTIONS.md](./docs/NAMING_CONVENTIONS.md) | Designing wire protocols, sync code, or SQL schemas. |
| [AI_PROVIDER_TYPES.md](./docs/AI_PROVIDER_TYPES.md) | Working on AI integration, adding providers, or modifying model selection
Code Local Notes
- `.claude/settings.json` blocks direct edits to lockfiles; use npm or pnpm commands instead.
- `.claude/settings.json` also auto-runs SDK lint autofix after TypeScript file edits.
- If you need reviewer
everything `anthropic`. That `anthropic` bucket includes mimo-named models the user ran *inside* Claude Code (e.g. `model=mimo-v2.5-pro`, logged in `~/.claude
interface-details` skills before writing surface code, reuse
existing components, themed tokens, and text primitives, and treat duplicate animation/UI primitives
as a review finding.
## Delegation shape (what worked)
- **Corridor-sized
impact revenue and user trust.
Treat every change like it's going through senior codereview:
- No lazy shortcuts or placeholder code
- Handle errors and edge cases properly
- Write code
when working with code in this repository.
## Project Overview
claude-reflect is a Claude Code plugin that implements a two-stage self-learning system:
1. **Capture Stage** (automatic): Hooks detect
AdapterError::not_supported())`
No existing files are modified beyond the registration points. Enforce via codereview.
## JSON Output Contract
Every command produces a response envelope:
```json
{
"version": "2.4",
"ok": true
agreement before correction.
Bad:
“Yes, you’re right, but…”
Good:
“No. The issue is…”
CodeReview Rules
When reviewing or modifying code:
- Do not assume the user’s diagnosis
Never: bug history, measurements, audit trails, review item numbers, dates, "PR #NNN", what an older commit did, or a restatement of the code. That belongs in the commit message
CLAUDE.md
This development guide applies to Codex, Claude Code, and OpenCode.
Repository and review rules are shared through:
@AGENTS.md
## Project Overview
E2B Infrastructure is the backend infrastructure powering E2B (e2b.dev
checker* to the external oracle, and how the review loop converges.
- [`external-oracle-process.md`](.claude/references/external-oracle-process.md) — running an independent frontier-model referee (Claude Code → another vendor's model via the Oracle CLI; current
SchemaMigrationPlan` (see `SchemaVersions.swift`). Each schema version must declare its own nested `@Model` snapshots; app code reaches the current version through top-level `typealias`es like `Skill = SchemaV1.Skill`. When adding
JSON.stringify(value ?? []);` then depend on `key`
- Memoize: `const stable = useMemo(() => value ?? [], [value]);`
- When reviewingcode, treat any `?? []`, `?? {}`, or inline `.map()/.filter()` in a dependency array as a bug.
## UI System
work logs
- `cc_sessions/agents/code-review.md` - Reviews implementations
- `cc_sessions/agents/context-refinement.md` - Updates context with discoveries
- `cc_sessions/agents/service-documentation.md` - Maintains service docs
### Supporting Infrastructure
- `cc_sessions/python/statusline.py` or `javascript/statusline.js` - Claude Code statusline with git info
A file Claude Code reads at the start of every session. It holds the commands, conventions and warnings the agent needs for this project.
Where does it go?
At the repository root. Claude Code also reads CLAUDE.md files in subdirectories when it works there.
What should it contain?
Build and test commands, the project's layout, conventions that aren't obvious from the code, and mistakes to avoid. Short files tend to work better than long ones.
CLAUDE.md or AGENTS.md?
Claude Code reads CLAUDE.md; most other agents read AGENTS.md. Many projects keep one and point the other at it.