Playground Concepts

The Playground is the dashboard view of a running Hyreflow session (recruit.hyreflow.ai/dashboard). It renders the plan the agent publishes, step status as work progresses, and the artifacts (CSVs) it produces.

The session model

Every task starts with a plan — a JSON array of short step labels — posted before any tool or credit call. The engine stores it; the Playground renders it as a live checklist.

hyreflow session start \
  --steps '["Recon: company + sizing","Pilot contact search","Approval gate","Full pull","Qualify + enrich","Deliver"]' \
  --user-prompt "find all senior PMs at fintech competitors"

As the agent works, it marks steps and emits sub-step updates:

hyreflow session update --index 0 --status running
hyreflow session update --index 0 --status completed
hyreflow session status --message "Found 8 competitor domains — checking each for open PM roles." --step-index 1

Step statuses: pending, running, completed, error, skipped. session status messages are plain-language narration — what's being searched, why, and what was found — not raw tool/payload/credit mechanics. See CLI Concepts for how --session-id is resolved when it's omitted, and what happens when a subcommand has no session to act on.

Outputs

Artifacts are registered by path reference, not contents — the engine stores where the CSV lives, the Playground links to it.

hyreflow session output --csv hyreflow/data/fintech-pms/fintech-pms.csv --label "Fintech competitor PMs"

Each output renders as a card with a Table and Shortlist view — the shortlist renders each row (company or candidate) as a card, with a fit-tier badge and evidence when the dataset has a fit_tier column, instead of a raw grid. Use Export HTML to download a standalone, styled shortlist page — a client-facing deliverable you can share without opening a spreadsheet.

Always write to hyreflow/data/<slug>/ and register the file. Never read a large CSV back into the agent's context or write to /tmp/ — the file-in/file-out convention is what keeps large runs (thousands of rows) cheap and inspectable.

Spend & ending a session

The Playground shows a spent pill — the credits every tool call made during this session has cost. Any credit call you run while the session is active counts toward it, and toward the session's spend cap if one is set.

A session starts with the workspace's default per-session cap, if you've set one. The Session Spending Limit modal — or session limit — raises, lowers, or removes the cap for this session alone, leaving the default and every other session untouched. When the run is done, close it:

hyreflow session end

Ending a session stops later calls from counting toward its spent total or cap, and clears it as the CLI's current session — so the next session start begins a clean run.

Approval gates

Credit-consuming and write actions (ATS pushes, bulk enrichment) sit behind a gate. The agent pilots a single row, surfaces the projected cost, and waits for approval before the full pull — visible as a paused step in the Playground.

Relationship to the CLI

The Playground is read-only narration; the CLI (or the agent driving it) is the writer. Everything you see in the UI was posted via session start / update / status / output. See CLI Concepts for the full command surface.