sourcegraph/CodeScaleBench

archive-run

Archive old completed benchmark runs to save disk space and speed up scans. Triggers on archive runs, clean up runs, disk space, old runs.

Quellsprache: Englisch

benchmark-audit

Audit benchmark suites against ABC framework (Task/Outcome/Reporting validity). Checks instruction quality, verifier correctness, reproducibility. Triggers on benchmark audit, audit benchmark, abc audit, task validity.

Quellsprache: Englisch

check-infra

Verify infrastructure readiness before launching benchmark runs — tokens, Docker, disk, credentials. Triggers on check infra, infrastructure check, ready to run, pre-run check.

Quellsprache: Englisch

compare-configs

Compare benchmark results across agent configurations (baseline, SG_full). Show where configs diverge. Triggers on compare configs, config comparison, which config wins, MCP impact.

Quellsprache: Englisch

cost-report

Token and cost analysis per run, suite, and config. Shows most expensive tasks and config cost comparison. Triggers on cost report, how much did it cost, token usage, spending.

Quellsprache: Englisch

generate-report

Generate the aggregate CSB evaluation report from completed Harbor runs. Triggers on generate report, eval report, ccb report, benchmark report.

Quellsprache: Englisch

ir-analysis

Compute information retrieval quality metrics (precision, recall, MRR, nDCG, MAP) comparing file retrieval across baseline and MCP configs against ground truth. Triggers on ir analysis, retrieval metrics, file recall, ground truth, search quality.

Quellsprache: Englisch

mcp-audit

Analyze MCP tool usage patterns, reward/time deltas conditioned on MCP adoption, and zero-MCP investigation. Triggers on mcp audit, mcp analysis, mcp impact, tool usage analysis, did mcp help.

Quellsprache: Englisch

mine-tasks

Point at any repo (GitHub, GitLab, Bitbucket, Azure DevOps, self-hosted, or local) and get eval tasks that compare baseline coding agents vs MCP-augmented agents. Mines merged PRs/MRs for real code-change tasks, auto-generates ground truth from patches,…

Quellsprache: Englisch

quick-rerun

Run a single benchmark task locally to verify a fix. Uses haiku for speed. Triggers on quick rerun, rerun task, verify fix, test task.

Quellsprache: Englisch

reextract-metrics

Batch re-extract task_metrics.json for all runs after extraction bug fixes or schema changes. Triggers on reextract metrics, refresh metrics, update task metrics, fix extraction.

Quellsprache: Englisch

repo-health

Run the repo health gate before commit or push to reduce doc drift and keep branches clean. Triggers on commit, push, before push, repo health, reduce drift, doc drift, clean branch.

Quellsprache: Englisch

run-benchmark

Configure and launch CodeScaleBench runs with current paired-run and curation guardrails.

Quellsprache: Englisch

run-status

Quick status check on the currently active benchmark run. Lighter than watch-benchmarks, scoped to recent activity. Triggers on run status, how's it going, are tasks done, active run.

Quellsprache: Englisch

scaffold-task

Scaffold a new Harbor-compatible benchmark task (SDLC or org-scale) and optionally a new benchmark suite. Generates task.toml, instruction.md, Dockerfile, test.sh, and registers the task. Triggers on scaffold task, new task, create task, add task, new…

Quellsprache: Englisch

score-tasks

Score individual task quality on instruction clarity, verifier quality, and reproducibility. Triggers on score task, task quality, rate tasks, task score.

Quellsprache: Englisch

sync-metadata

Reconcile task metadata between selected_benchmark_tasks.json and task.toml files. Finds and fixes drift. Triggers on sync metadata, check metadata, metadata mismatch, reconcile tasks.

Quellsprache: Englisch

triage-failure

Investigate a specific failed benchmark task — read logs, identify root cause, check if known pattern, suggest fix. Triggers on triage, investigate failure, debug task, diagnose failure.

Quellsprache: Englisch

validate-tasks

Pre-flight validation of benchmark tasks before launching runs. Catches truncated instructions, metadata mismatches, missing test scripts. Triggers on validate tasks, preflight, pre-flight check, check tasks.

Quellsprache: Englisch

watch-benchmarks

Live benchmark monitoring — scan run directories, classify task status, fingerprint errors, present structured summaries. Triggers on watch benchmarks, benchmark status, run status, monitor runs.

Quellsprache: Englisch

whats-next

Analyze current benchmark state and recommend what to work on next. Triggers on whats next, what should I do, next steps, prioritize work.

Quellsprache: Englisch

audit

Run repo health checks, validate benchmark tasks, and audit run integrity.

Quellsprache: Englisch

evaluate

Extract metrics, score traces, and evaluate benchmark task results.

Quellsprache: Englisch

infra

Check infrastructure readiness, manage MCP tools, and audit system dependencies.

Quellsprache: Englisch

next

Plan upcoming work, analyze coverage gaps, and recommend next steps for benchmarking.

Quellsprache: Englisch

report

Generate evaluation reports, analyze run costs, and compare configurations.

Quellsprache: Englisch

run

Launch and manage CodeScaleBench benchmark runs with paired-run guardrails, quick reruns, and execution orchestration.

Quellsprache: Englisch

scaffold

Create, mine, and validate new benchmark tasks and task suites.

Quellsprache: Englisch

status

Monitor active runs, check task completion status, and watch benchmark execution progress.

Quellsprache: Englisch

triage

Investigate and triage failed benchmark tasks, analyze root causes, and plan reruns.

Quellsprache: Englisch