broomva/broomva.tech/broomva-cli

broomva — CLI Quick Reference

Rust CLI at ~/.cargo/bin/broomva, source at crates/broomva-cli/ in the broomva.tech monorepo. Talks to broomva.tech/api/* (Phase 1 telemetry + existing prompts/skills/context endpoints).

Standard commands

broomva prompts list [--mine] [--category X] [--metrics] [--sort skill_invokes]
broomva prompts get <slug> [--raw|--json]
broomva prompts pull <slug> [-o file.md] [--json]
broomva prompts push file.md [--create]
broomva skills list [--layer X]
broomva context conventions
broomva auth status

Running prompts (auto-traced)

This is the contract that turns every prompt invocation into a typed row server-side. Follow these five steps when the user asks for a prompt, or when you invoke a prompt as instructions during a task.

1. Set the source flag once per session:

``bash export BROOMVA_SOURCE=skill ``

1. Pull the prompt:

``bash broomva prompts pull <slug> --json 2>&1 | tee /tmp/broomva-last.json ` The CLI writes the prompt to <slug>.md (or -o path) and emits a JSON line on stderr: {invocation_id, prompt_slug, prompt_version, posted}`.

1. Use the prompt body in your reply / as the system instructions

for the work the user asked you to do.

1. MANDATORY: After completing the work the prompt asked for, call:

``bash broomva prompts complete <invocation_id> \ --status completed \ --model <your-model-name> \ --latency-ms <wall-clock-ms> \ --tokens-in <input-tokens> \ --tokens-out <output-tokens> ` Use --status failed if the work could not be completed and include --error-message`. This is how prompts get evaluated. Skipping it breaks the feedback loop.

1. Optional: capture explicit feedback if the user gave a clear

thumbs up/down: ``bash broomva prompts feedback <invocation_id> --slug <slug> --signal up --text "user's words" ``

Environment

API routes the CLI talks to

• GET /api/prompts[?include=metrics&sort=...] — list (Phase 2 ?include=metrics)
• GET /api/prompts/:slug[?include=metrics] — get
• POST /api/prompts — create (auth required)
• PUT /api/prompts/:slug — update (auth required)
• DELETE /api/prompts/:slug — delete (auth required)
• POST /api/invocations — telemetry beacon (anonymous-OK, rate-limited per IP/user)
• PATCH /api/invocations/:id — completion update (anonymous-OK if row has no user_id)
• POST /api/feedback — explicit user signal (anonymous-OK)
• GET /api/feedback?prompt_slug=...&limit=... — recent feedback
• GET /api/metrics/overview?since=24h|7d|30d|all — hero KPI strip
• GET /api/metrics/runs?prompt_slug=&source=&limit=&before= — paginated runs
• GET /api/metrics/volume?bucket=hour|day&since=24h|7d|30d — timeseries
• GET /api/metrics/prompts/:slug — per-prompt strip
• GET /api/skills / GET /api/skills/:slug — skills roster
• GET /api/context — project context + conventions

Auth

Token resolution order: --token flag > BROOMVA_TOKEN env > ~/.broomva/config.json. Run broomva auth login for the device-code flow, or broomva auth login --manual to paste a token.

Source attribution

When this skill invokes the CLI inside Claude Code, it sets BROOMVA_SOURCE=skill before the pull, so the row carries source='skill' server-side. Terminal users default to cli. Programmatic callers wrapping the CLI should set BROOMVA_SOURCE=api.