Add CLI Agent Benchmark Tasks documentation#182
Open
oddessentials wants to merge 1 commit intomainfrom
Open
Conversation
Define a structured benchmark suite for evaluating AI-backed CLI tools against this polyglot codebase. Covers cross-service schema evolution, bug detection, dependency upgrades, test authoring, self-review, feature planning, linter compliance, health check standardization, and architecture comprehension. Each task has a deterministic grading rubric with automatable pass/fail criteria (106 total points across 10 tasks). https://claude.ai/code/session_018FhKabQMCUFrhJq8TGS8dm
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
This PR introduces a comprehensive benchmark suite for evaluating AI-backed CLI tools against the Distributed Task Observatory polyglot codebase. The document defines 10 deterministic tasks spanning multiple difficulty levels and capability areas, each with detailed grading rubrics and automated verification criteria.
Key Changes
docs/CLI_BENCHMARK_TASKS.md(408 lines)Notable Implementation Details
Grading Automation: Most rubric items (>80%) are fully automatable via shell commands, Python assertions, or exit-code checks. Only 4 items require manual human review (self-review quality, plan accuracy, error-free explanations).
Real Bug Catalog: Task 2 includes a table of 5 actual bugs seeded in the codebase for agents to discover and fix, with specific line numbers and verification methods.
Multi-Language Coverage: Tasks exercise TypeScript (Gateway), Python (Processor), Go (Metrics Engine, Read Model), and Rust (Web PTY Server, TUI) services.
Progressive Difficulty: Tasks range from Easy (linting, architecture explanation) to Hard (cross-service schema evolution, end-to-end feature implementation), allowing benchmarking of agents at different capability levels.
Deterministic Environment: All tasks run in a clean checkout with specified tool versions (Node 20, Python 3.11, Go 1.21, Rust 1.83) and wall-clock time limits (10–30 minutes).
Purpose
This benchmark enables objective, reproducible evaluation of AI CLI agents on realistic polyglot codebase tasks, with clear success criteria and automated grading where possible.
https://claude.ai/code/session_018FhKabQMCUFrhJq8TGS8dm