メニュー
HoneyHive CLI install, discovery, and usage reference. Shared by honeyhive-instrument, honeyhive-evaluate, and honeyhive-improve skills. Use to install the CLI, discover commands, inspect schemas, and file doc-gap issues.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Instrument an AI application with HoneyHive tracing. Use when the user wants to add observability, capture traces, set up the SDK, configure exporters, or wire OpenTelemetry-compatible instrumentation for an agent, RAG pipeline, or LLM application. Triggers on phrases like "add tracing", "instrument my app", "wire up HoneyHive", "set up the SDK", "capture spans", "OTEL exporter", or any mention of `HoneyHiveTracer.init`, `@honeyhive/api-client`, `client.sessions.startSession`, `client.events.createEvent`, or `OpenInference`/`Traceloop` instrumentors targeting HoneyHive.
Decode a HoneyHive Discover URL, query the flagged sessions and their trace trees, classify each finding as a true or false positive with evidence, and recommend specific guardrails (hooks, evaluator changes, prompt additions) to prevent recurrence.
Set up and verify HoneyHive experiments for an AI agent or LLM workflow. Use when the user wants to evaluate an agent, run an experiment, create or reuse a dataset, add evaluators or rubrics, compare experiment runs, or wire `evaluate()` with HoneyHive. Triggers on phrases like "evaluate my agent", "run an experiment", "create an eval dataset", "add evaluators", "compare runs", `honeyhive experiments`, `honeyhive datasets`, or `evaluate()`.
Debug and improve an AI agent or LLM workflow using HoneyHive trace data. Use when the user has a failing session, event, evaluator, experiment run, alert, or vague production complaint and wants root-cause analysis, a minimal fix plan, or a verified improvement loop. Triggers on phrases like "debug this trace", "investigate a failing session", "why did this run fail", "root cause", "production issue", `honeyhive events search`, `honeyhive experiments`, or evaluator failure analysis.
| name | honeyhive-cli |
| description | HoneyHive CLI install, discovery, and usage reference. Shared by honeyhive-instrument, honeyhive-evaluate, and honeyhive-improve skills. Use to install the CLI, discover commands, inspect schemas, and file doc-gap issues. |
| license | MIT |
| metadata | {"version":"0.1.0","homepage":"https://docs.honeyhive.ai"} |
brew tap honeyhiveai/tap && brew install honeyhivecurl -fsSL https://github.com/honeyhiveai/honeyhive-cli/releases/latest/download/install.sh | sh (requires glibc — works on Ubuntu, Debian, Fedora, RHEL, WSL; does not work on Alpine/musl)honeyhive --help — top-level namespaces and global flagshoneyhive <namespace> --help — subcommands (e.g. honeyhive events --help)When a command accepts file input or structured arguments:
honeyhive <command> --show-file-schema — JSON schema for file-based inputhoneyhive <command> --show-argument-schema <argument-name> — JSON schema for a specific structured argument (use the argument name without the -- prefix)Use these instead of guessing flags or payload shapes.
Before any skill that depends on HoneyHive, confirm the CLI is working and credentials are valid. Run these checks in order; stop on first failure.
honeyhive --help must succeed. If not, install per the instructions above.HH_API_KEY and HH_DATA_PLANE_URL must be in the environment (never hardcoded in source). If missing, direct the user to their project settings page to find both values: https://app.us.honeyhive.ai/settings/project/keys for multi-tenant, or https://app.<tenant>.us.honeyhive.ai/settings/project/keys for dedicated deployments (see references/dedicated-deployments.md).honeyhive events search --filters '[]' --limit 1 --page 1
{"events":[...]}) = good. An error or auth failure = stop and surface. If the CLI is not installed, fall back to curl — see references/network-validation.md for the exact command and status-code table.
On failure at any step, surface the exact error and stop. Do not proceed into skill-specific work with broken credentials or connectivity.The sanity check produces: status (pass/fail), hh_data_plane_url, deployment_type, and probe_method. The calling skill decides whether and where to persist these (e.g. honeyhive-instrument writes state/network-validation.json; other skills may not need state files).
When a skill hits missing CLI coverage, incomplete docs, or an unsupported framework/integration, file automatically:
gh issue create --repo honeyhiveai/skills \
--title "Doc gap: <short description>" \
--body "**Skill:** <skill-name>
**CLI/SDK version:** <detected version>
**Framework:** <framework and version>
**Gap:** <what is missing>
**Workaround:** <what was done instead>"
Report the created issue URL to the user.