Skip to content

AGENTS.md

Latest commit

 

History

History
35 lines (26 loc) · 2.22 KB

AGENTS.md

File metadata and controls

35 lines (26 loc) · 2.22 KB

AGENTS.md

This repository uses a multi-agent coding system. All implementer agents must follow the project skills defined in .opencode/skills/.

Skills

Skill Location Purpose
project-coding .opencode/skills/project-coding/SKILL.md Coding conventions, naming, architecture patterns, error handling
project-build .opencode/skills/project-build/SKILL.md Build commands, prerequisites, environment setup
project-test .opencode/skills/project-test/SKILL.md Test framework, conventions, patterns, coverage
project-documentation .opencode/skills/project-documentation/SKILL.md TSDoc, README standards, API docs
project-security .opencode/skills/project-security/SKILL.md Input validation, path traversal, dependency security
project-code-examples .opencode/skills/project-code-examples/SKILL.md Catalog of reusable code patterns

Code Examples

Concrete, annotated snippets extracted from the real codebase live in .code-examples-for-ai/. Read these before implementing a feature that uses one of the documented patterns.

Stack

  • Runtime: Bun 1.x (TypeScript executed natively — no build step)
  • Language: TypeScript strict mode (ESNext, verbatimModuleSyntax, noUncheckedIndexedAccess, exactOptionalPropertyTypes)
  • Analysis parser: web-tree-sitter (multi-language) for graph/watch analysis (replaces @typescript-eslint/typescript-estree TypeScript-only parsing)
  • Schema validation: Zod 4.x
  • Test framework: Vitest 4.x (unit in tests/, E2E in e2e/ via Docker)
  • Architecture: CLI command layer (src/commands/) + core services (src/cache/, src/files/, src/http/, src/search/, src/platform/) + analysis engine (src/analysis/ — multi-language Tree-sitter symbol extraction, dependency graph, PageRank)

Key Rules

  • All recoverable errors use Result<T, E> — never throw across command boundaries
  • All external JSON is validated with Zod safeParse() before use
  • All subject strings and tracked file paths are validated against traversal guards before filesystem use
  • Tests must cover happy path + every ErrorCode branch per command
  • No TODOs, placeholder logic, or commented-out dead code in production paths