manage-experiments
Create, update (metrics, dates, description, hypothesis, baseline/comparison treatments, status), or delete Harness FME feature-flag experiments. Delegates metric selection to choose-metric, metric creation to create-metric, event wiring to instrument-metric, results to review-experiment-results, flag/definition prep to create-feature-flag and update-flag-targeting. Use when asked to create, modify, pause, resume, complete, or delete an experiment. Do not use for explaining results (review-experiment-results). Trigger phrases: create experiment, set up A/B test, design experiment, launch test, update experiment, pause/resume/complete experiment, delete experiment.
Harness · 已有静态检查记录
安装与使用
运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。
npx skills add harness/harness-skills --skill manage-experiments技能说明
name: manage-experiments description: >- Create, update (metrics, dates, description, hypothesis, baseline/comparison treatments, status), or delete Harness FME feature-flag experiments. Delegates metric selection to choose-metric, metric creation to create-metric, event wiring to instrument-metric, results to review-experiment-results, flag/definition prep to create-feature-flag and update-flag-targeting. Use when asked to create, modify, pause, resume, complete, or delete an experiment. Do not use for explaining results (review-experiment-results). Trigger phrases: create experiment, set up A/B test, design experiment, launch test, update experiment, pause/resume/complete experiment, delete experiment. metadata: author: Harness version: 2.1.0 mcp-server: harness-mcp license: Apache-2.0 compatibility: Requires the Harness MCP server or the Harness CLI
Manage Experiments
Create, update, or delete FEATURE_FLAG experiments. AI_CONFIG mutation workflows are not implemented here; use read-only results inspection or a separately supported workflow. Orchestrates decisions and delegates metric selection, flag setup, and results interpretation to specialist skills.
Related: choose-metric (metric selection), create-metric (metric creation), instrument-metric (event wiring), review-experiment-results (results), create-feature-flag, update-flag-targeting.
Tools
Works through the Harness MCP server or the Harness CLI; names are from tool-map.md.
| Operation | MCP | CLI |
|---|---|---|
| List experiments | harness_list · fme_experiment · filters: { parent_type, parent_name?, environment_id?, status?: ["ACTIVE", "PAUSED"], … } · compact: false | harness list experiment --parent-type FEATURE_FLAG [--search <name>] [--status ACTIVE], then again with --status PAUSED |
| Get experiment | harness_get · fme_experiment · params: { experiment_id } | harness get experiment <experiment-id> |
| Create experiment | harness_create · fme_experiment · params: { environment_id } · body: { parent, name, startAt, endAt, baselineTreatment, comparisonTreatments, rule: "default rule", owners?, … } | harness create experiment <name> --env <env-id> -f experiment.json --json |
| Update experiment | harness_update · fme_experiment · params: { experiment_id } · body: { description?, hypothesis?, startAt?, endAt?, keyMetrics?, status?, rule?, … } | harness update experiment <experiment-id> --set description=foo --set status=PAUSED |
| Delete experiment | harness_delete · fme_experiment · params: { experiment_id } | harness delete experiment <experiment-id> |
| List environments | harness_list · fme_environment · compact: false | harness list fme_environment --json |
| Get definition | harness_get · fme_feature_flag_definition · params: { feature_flag_name, environment_id } | harness get feature_flag:definition <flag-name> --env <env-id> |
Instructions
Phase 1: Establish scope
Follow scope-establishment.md.
Phase 2: Choose operation
Create: new experiment on a feature flag.
Update: change description, hypothesis, dates, metrics, baseline/comparison
treatments, or status (pause/resume/complete/archive).
Delete: hard delete (permanent, no archive/restore).
Create experiment
Step 1: Resolve parent and environment
Default parent type: FEATURE_FLAG. If flag doesn't exist, route to create-feature-flag, then return here. List environments; ask which environment if not stated.
Stop for AI_CONFIG or CONFIG mutations. The declared tools cannot verify their parent definitions, treatment names or traffic readiness. Do not substitute a feature-flag lookup or user guesses for that check. Read-only experiment/results inspection remains possible; ask for a supported AI Config workflow before writing.
Step 2: Confirm treatments against live definition
Get definition for the flag in the chosen environment. 404 = no definition; route to update-flag-targeting to initialize it, then return here.
Read treatments[].name. Experiment's baselineTreatment and comparisonTreatments must match real treatment names. Default baseline = definition's baselineTreatment, but confirm with user. Fewer than 2 treatments → stop (see stop-conditions.md).
Check traffic readiness: isKilled: false, trafficAllocation > 0, baseline and each comparison treatment have size > 0 in defaultRule buckets (or in the targeting rule the experiment will use). If any check fails, warn and ask whether to proceed. Fixing it requires update-flag-targeting.
List experiments for the parent flag and environment, with ACTIVE and PAUSED status. Apply the experiment check: ACTIVE means confirm the user wants a second experiment; PAUSED means warn and require explicit acknowledgement.
Step 3: Hypothesis
Get hypothesis: "if [change], then [metric] will increase/decrease because [reasoning]" (max 500 chars). Link concepts.md. Missing or non-causal hypothesis → stop (see stop-conditions.md).
Step 4: Metrics
Hand off to choose-metric to select key and supporting metrics. If no suitable metric exists, route to create-metric (and instrument-metric if event missing). Return here with metric IDs o
阅读完整来源 ↗