Vibeleaderboard
Index / tool
Visit github.com
Category
AI Agents
Rank
No. 1557Tools index
Pricing
Open Source
Type
TOOL
Builder
gsd-build
GitHub
52 stars
Date

About

File-based context resolution for AI agent DAG workflows. Three primitives, zero dependencies.

What it does

Context Packet coordinates multi-step agent pipelines described as graphs. Each completed step records a packet containing status, summary, body, and optional structured data. Later steps collect transitive upstream packets, combine graph and node instructions, fit content to a token budget, and receive a semantic hash for change detection.

Why it's ranked here

The design is small, inspectable, and useful for local agent pipelines. It validates graph structure, prioritizes nearby context under tight budgets, and supports library, command-line, and MCP workflows. The implementation also has meaningful rough edges: token counts are estimated, packet writes can replace prior results, and the package includes an MCP dependency despite claiming none.

What's good

Execution-order edges and data-consumption edges remain distinct in the graph model. Cycle and unknown-node checks catch malformed pipelines early. Context resolution preserves summaries before trimming bodies, prioritizes direct dependencies, reports missing inputs, and marks truncation. Canonical SHA-256 hashes exclude timestamps, so unchanged semantic inputs produce stable identifiers. Temporary-file writes, synchronization, and renaming reduce partial-write risk.

Tradeoffs

The token budget uses four characters per token, so actual model usage can differ. Delimiter wrapping labels upstream text as data but does not sanitize or remove hostile content. Re-submitting a node overwrites its packet, despite documentation describing packets as immutable. The runner uses a basic command splitter, captures complete output in memory, and stops only after every parallel task in a failing level finishes.

How to use it well

Use it for local, reproducible pipelines where several agents or scripts exchange bounded text through a known dependency graph. Choose the command runner for stdin-based tools, or MCP when an agent needs broader capabilities between reading context and submitting work. Treat hashes as skip signals and packet files as inspectable state. It does not supply the agent, hosted coordination, or database-backed workflow state.

Technical notes+

The package is strict TypeScript targeting ES2022 with NodeNext modules, configured in tsconfig.json; package.json requires Node 18 or newer, publishes dist, exposes an ESM library and CLI binary, and declares @modelcontextprotocol/sdk as a runtime dependency. src/resolve.ts traverses both dependency kinds, estimates tokens from character count, mutates returned packet bodies when partially truncating, and builds delimiter-wrapped prompts through src/sanitize.ts. src/hasher.ts recursively sorts object keys and hashes summaries, bodies, and optional data. src/store.ts writes through temporary files, fsync, and rename, but does not prevent replacement. src/runner.ts executes each topological level with Promise.all. src/graph.ts attempts optional YAML loading with require("js-yaml"), while the package is ESM and does not declare js-yaml. Tests in src/test/context-packet.test.ts cover initialization, dependency enforcement, resolution, truncation, status, hashing, cycles, and delimiter wrapping.

Observed

License
MIT
Primary language
TypeScript
Install surface
npm package named context-packet
Runtime
Node.js 18 or newer, ES modules
Interfaces
ESM library, command-line interface, and stdio MCP server
Published package contents
The package manifest includes only the compiled dist directory
Runtime dependencies
Declares @modelcontextprotocol/sdk
Tests
Repository includes Node test-runner coverage under src/test

Read from README.md, package.json, src/cli.ts, src/index.ts, src/graph.ts, src/store.ts, src/types.ts, src/hasher.ts, src/runner.ts, src/resolve.ts, src/sanitize.ts, src/mcp-server.ts, src/bin/context-packet.ts, src/test/context-packet.test.ts, tsconfig.json.

What it can do

  • Resolve file-based context for AI workflows

    Context files and workflow definitionsResolved context data for AI agents

  • Build directed acyclic graph (DAG) workflows

    Workflow specifications and agent definitionsExecutable DAG workflow structure

  • Process context packets between workflow nodes

    Context packets and node configurationsTransformed context data

  • Execute AI agent workflows

    DAG workflow and input contextWorkflow execution results

  • Parse and structure context files

    Raw context filesStructured context objects

Tags

ai-agentscontextdagworkflowstypescript

Tech Stack

Node.jsTypeScript

Comments (0)

No comments yet

Editorially curated, with community endorsements as a secondary signal. Corrections welcome.