# Dataset Workflows Skill Use when managing AgentClash datasets via CLI — create versions, import/export examples, run evals, CI gates, synthetic generation, trace import, candidate review, and regression suite sync. Source: https://www.agentclash.dev/docs/agent-skills/agentclash-dataset-workflows Markdown export: https://www.agentclash.dev/md/docs/agent-skills/agentclash-dataset-workflows Canonical source: `web/content/agent-skills/agentclash-dataset-workflows/SKILL.md` Markdown export: `/docs-md/agent-skills/agentclash-dataset-workflows` ## Use This Skill When Use when managing AgentClash datasets via CLI — create versions, import/export examples, run evals, CI gates, synthetic generation, trace import, candidate review, and regression suite sync. ## Full SKILL.md ````markdown --- name: agentclash-dataset-workflows description: Use when managing AgentClash datasets via CLI — create versions, import/export examples, run evals, CI gates, synthetic generation, trace import, candidate review, and regression suite sync. metadata: agentclash.role: dataset agentclash.version: "1" agentclash.requires_cli: "true" --- # AgentClash Dataset Workflows ## Purpose End-to-end dataset operations in a workspace: versioned example banks, eval runs against challenge packs, CI gating, synthetic generation, production trace import, and promotion into regression suites. ## Use When - Building or curating labeled examples for prompt or agent evals. - Gating merges on dataset eval pass rate vs a baseline. - Importing OTEL/Braintrust/LangSmith/Phoenix/AgentClash traces as reviewable candidates. - Syncing a dataset version into a linked regression suite. ## Do Not Use When - The user only needs a one-off challenge-pack run — use `agentclash-eval-runner`. - Prompt matrix experiments without a dataset artifact — use `agentclash-prompt-eval-playground`. - Harness coding tasks — use `agentclash-agent-harness-setup`. ## Inputs Needed - Workspace ID and dataset ID (create datasets via API/UI if none exist). - For eval/gate: dataset version ID, challenge pack version ID, challenge key, deployment IDs. - For gate: baseline ID and candidate run ID (or `--eval` to start eval inline). - For generate: `--count`, `--provider-account`, `--model`. ## Environment ```bash export AGENTCLASH_API_URL="https://api.agentclash.dev" agentclash workspace use agentclash dataset list agentclash dataset view --json ``` ## Procedure 1. Inspect dataset and versions (`list`, `view`, `version list`). 2. Import or export examples; optionally create a version snapshot. 3. Run a dataset eval or attach an existing run. 4. Gate with `dataset test` against a baseline (CI-friendly `--format junit`). 5. Optionally generate synthetic examples, import traces, promote candidates, sync regression suite. ## Commands ### Inspect and mutate examples ```bash agentclash dataset list agentclash dataset view agentclash dataset version list agentclash dataset version create --label "v2-seeds" agentclash dataset import examples.jsonl agentclash dataset export --version > out.jsonl agentclash dataset example list agentclash dataset example add --input '{"messages":[...]}' --expected '{"score":1}' agentclash dataset example edit

--expected '{"score":1}' agentclash dataset example rm

``` ### Eval and CI gate ```bash agentclash dataset eval \ --version \ --pack \ --challenge \ --deployment agentclash dataset test \ --baseline \ --run \ --min-pass-rate 0.9 \ --max-regressions 0 \ --format junit # Start eval then gate in one command agentclash dataset test \ --eval \ --version \ --pack \ --challenge \ --deployment \ --baseline \ --timeout 30m ``` ### Synthetic generation ```bash agentclash dataset generate \ --count 50 \ --provider-account \ --model \ --create-version \ --version-label "synthetic-v1" \ --follow ``` ### Trace import and promotion ```bash agentclash dataset import-traces traces.json --source otel agentclash dataset import-traces --source agentclash --run --run-agent agentclash dataset trace-candidates list --status pending agentclash dataset promote

--tag production --expected '{"score":1}' ``` ### Regression suite sync ```bash agentclash dataset sync-regression-suite \ --version \ --pack \ --challenge \ --suite-name "Dataset regression bank" ``` ## Expected Output - Eval creates a run; gate returns pass/fail with regression counts. - `dataset test --format junit` exits 0 on pass, 1 on gate failure (422). - Generate with `--follow` polls job until completion. ## Failure Modes - Gate without `--baseline` → required. - Gate without `--run` and without `--eval` → provide one. - Generate missing provider/model → all three of count, provider-account, and model required. - Sync regression without version/pack/challenge → all three flags required. ## Safety Notes - Trace imports may contain production data — apply `--redaction` JSON when importing sensitive metadata. - Baseline comparisons affect release gates — confirm baseline ID before CI integration. - Exported JSONL may include prompts with secrets — scrub before sharing externally. ## Report Back Format ```text Dataset: Version: Eval run: Gate: — pass rate , regressions Candidates: Regression suite: Next: agentclash run scorecard ``` ## Related Skills - `agentclash-hub` - `agentclash-eval-runner` - `agentclash-regression-flywheel` - `agentclash-ci-release-gate` - `agentclash-scorecard-reader` - `agentclash-prompt-eval-playground` ## Related Docs - `/docs-md/guides/datasets-overview` - `/docs-md/guides/ci-cd-workload-recipes` - `/docs-md/reference/cli` ````