Skip to content

AGENTS.md

Latest commit

 

History

History
81 lines (69 loc) · 4.79 KB

AGENTS.md

File metadata and controls

81 lines (69 loc) · 4.79 KB
Raw

Agent Operating Guide

Build and maintain tv, one Rust-native CLI for TradingView Desktop automation and explicit Desktop-free data and account operations. This is contributor guidance; release users get packaging/agent/AGENTS.md.

Project boundaries

  • Keep data sources and side effects explicit. Desktop-free requests must not silently become Desktop operations. Preserve the shared JSON envelope, documented source semantics, and practical information used by consumers.
  • Assess compatibility against the current public CLI, JSON, and Rust APIs and actual downstream use. The old JavaScript command migration is closed; newly found legacy commands need a concrete workflow benefit before becoming work items. Historical migration notes explain decisions, not automatic scope.
  • Keep private downstream orchestration outside the core CLI unless a concrete consumer demonstrates that it belongs here. MCP server work, cookie/session import/export, trading bots, and broad generic UI expansion are not planned.
  • Keep account-local IDs, credentials, raw live payloads, and machine-specific absolute paths out of tracked files. Use synthetic examples and public-safe evidence; see development guidance.

Read for the task

Start with the affected code and its consumers. Use this table when context is needed; it is not a mandatory reading sequence.

Task Reference
Product purpose and usage README
Select or resume planned work Current work
Data sources, JSON meaning, side effects Source taxonomy
Crate ownership or dependency boundaries Architecture
Public library contracts Rust API
Choose checks or investigate implementation conventions Development
TradingView internal behavior Internal APIs and the affected adapter
Prepare distribution Release packaging
Plan a complex feature or significant refactor Planning guidance
Operate the binary through an agent Runtime guide and the relevant runtime skill

Code, current contracts, and explicit decisions take precedence over historical plans. When they disagree, investigate and correct the affected current source within scope. Do not turn a documentation disagreement into permission to edit unrelated work. Mark unresolved consequential facts UNCONFIRMED.

Execution and verification

Complete the authorized outcome, including affected callers, documentation, checks, and related fixes. Use the validation guidance to select checks. Ordinary deterministic checks use fixtures; live smokes are separate, opt-in operations with their own targets, limits, and authority. Reuse valid evidence when its inputs have not changed. Distinguish source inspection, fixture execution, and actual Desktop/provider observations.

Keep the detailed state of a substantial change in its existing work record. A roadmap owns direction, an inventory owns priority, and an ExecPlan owns the change's decisions and acceptance evidence. The optional continuity skill prepares a handoff only when explicitly requested; it is not an every-turn task or a source of new PM, commit, or delegation authority.

Repository conventions

  • crates/cli owns the binary and operation adapters; core owns envelopes and errors; model owns I/O-free interpretation and shaping. market, scanner, and pine own credential-free Desktop-free services; mcp owns the internal authenticated official-MCP client; cdp owns Desktop transport.
  • Prefer English for repository docs. Use descriptive work names rather than ordinal aliases. Public docs explain usage, reproduction, and maintenance.
  • Commit related authorized changes in coherent batches using the commit convention. Preserve staged and concurrent work; follow the assigned PM/implementer commit boundary.
  • Never push unless explicitly requested in the current turn. Release work keeps versioning, notes, packaging, and CI fixes separate from feature work.
  • CLAUDE.md shares this guide. Runtime skill sources live in skills/; .agents/skills/<name> links to them, and .claude/skills shares that root. Edit runtime skills only in skills/. Contributor skills remain in .agents/skills; distribution uses an explicit runtime allowlist and real files.
  • Current work lives in docs/plans; completed plans in docs/plans/archives; research and historical rationale in docs/notes. Do not rewrite frozen evidence merely to match a new documentation style.