grapha-dataflow

name	grapha-dataflow
description	Use when work in this repo needs forward dataflow traces to terminals, reverse traces to entry points, or entry-point listings.

Grapha — dataflow tracing

Tool map

Task	MCP tool	CLI fallback
Forward trace: from symbol to terminals	mcp__grapha__trace (direction: "forward")	grapha flow trace <symbol>
Reverse trace: from symbol back to entry points	mcp__grapha__trace (direction: "reverse")	grapha flow trace <symbol> --direction reverse
List auto-detected entry points	(use search with role: "entry_point")	grapha flow entries

When to run each

• Forward trace when you need to know what side effects a symbol's data eventually causes — does this view-model write to the network, the local store, a cache, or fire an event? Useful before adding side-effecting code or auditing a privacy/security flow.
• Reverse trace when you need to know who reaches a symbol — which screens, jobs, or routes ultimately hit this function? Useful for impact analysis at the entry-point level (more user-facing than get_impact's symbol-graph view).
• Entry-point listing when orienting in a large project — what are the user-facing surfaces (SwiftUI Views, @Observables, fn main, route handlers)?

Recipes

• Tight forward trace: mcp__grapha__trace({ symbol: "<sym>", direction: "forward", depth: 5 }). Depth defaults to 10 forward / unlimited reverse — start tighter and widen only if you don't see the terminal you expected.
• Reverse-from-terminal: pick a sensitive sink (e.g., the network call), reverse-trace, and the result is the set of entry points that can hit it. Pair with grapha-quality's get_impact for a fuller change-risk picture.
• For CLI flows with long traces, pipe Grapha JSON through FACE for grouping and paging. For MCP-only sessions, use Grapha's built-in cluster: true and page through bands.

Tips

• Terminals are auto-classified (network, persistence, cache, event) — see the project's grapha.toml for any user-defined classifier rules.
• If a forward trace ends suspiciously early, the missing edge is often a cross-module call that's ambiguous in tree-sitter mode; a fresh grapha index . (then mcp__grapha__reload) often resolves it because index-store extraction adds USR-precise edges.
• Forward and reverse traces are not symmetric — forward follows data, reverse follows reachability. Pick the one that answers your actual question.