mcpField
P
pod-sales

pod-sales

github.com/ruvnetMarketing & growth80.0k58 installsUpdated today
Repository

About pod-sales

pod-sales provides a single-shot skill to run, demonstrate, or schedule one iteration of the sales pod without launching real LLM workers. It validates the pod template, reserves a budget, and records a status envelope; live execution is refused in this build.

Key features

  1. 01Validates pod templates and agent types
  2. 02Reserves budget in a file ledger
  3. 03Dry-run without invoking a model
  4. 04Writes pod status to backing store
  5. 05Emits a JSON result line

Use cases

  1. 01Smoke-testing a sales pod iteration
  2. 02Demonstrating the sales autopilot
  3. 03Scheduling a dry-run sales tick

Install

Copy the command and run it in a terminal.

$npx skills add https://github.com/ruvnet/ruflo --skill pod-sales

After installing, type/pod-salesin Claude Code to confirm it loaded.

Details

From the project README

Surfaces pod-tick.mjs as a single-shot skill for the sales pod. Use when Claude Code needs to demonstrate, smoke-test, or schedule one iteration of the sales autopilot without spawning real LLM workers.

Algorithm

Implementation: scripts/pod-tick.mjs.

  1. Parse args (--pod-template, --base-path, --dry-run / --live, --budget-cap-usd, --tick-id). --live is refused with exit code 3 in Phase 2.
  2. Load the pod template JSON (default: templates/sales.json).
  3. Validate via validatePodTemplate(json) — schema in v3/@claude-flow/cli/src/business-pods/pod-schema.ts and inlined in pod-tick.mjs so the script runs without a built CLI. Throws with a JSON-pointer path on the first violation.
  4. Resolve every agent.agentType against KNOWN_AGENT_TYPES. Unknown types abort with exit code 2 and an actionable error.
  5. Reserve min(budgetUsdPerRun, --budget-cap-usd) USD against the file-based ledger at <base-path>/budget/<roomId>.json. Honors reservationExpiryMs (default 60_000 ms, bounded to [5000, 300000] per ADR-164.1 §3.2). TODO(adr-164.1): swap the file ledger for the atomic SQLite tracker in Phase 3.
  6. Build per-agent prompts (kickoff scoped to the bench description + pod's PII policy). In --dry-run they are logged to stderr and the model is never invoked.
  7. Commit the reservation. Dry-run actual = $0; live actual = the reserved amount (Phase 3 wires real claude -p --max-budget-usd reporting).
  8. Append a pod-status envelope to the Phase-1 backing store at <base-path>/.agentbbs/room-<derivedRoomId>.jsonl so subsequent federation_bbs_watch calls see the tick.
  9. Emit a single JSON line on stdout: {podName, tickId, agentsRan, totalUsd, envelopeId, status}.

Exit codes

  • 0 — tick succeeded (status === 'success')
  • 2 — invalid template / unknown agent type / budget exhausted / arg error
  • 3 — --live requested (refused in Phase 2)

Phase 3 surfaces (not in this build)

  • --live mode: dispatch each prompt through claude -p headless or a Managed Agent and capture the actual --max-budget-usd reported spend.
  • Atomic SQLite budget tracker (ADR-164.1 §3.2).
  • Multi-pod variants (pod-marketing, pod-finance, ...) — each gets its own skill once Phase 3 ships.

Comments

Markdown supported