Skillget

dont-burn-it-all

skill
Security Audit
Warn
Health Warn
  • License — License: MIT
  • Description — Repository has a description
  • Active repo — Last push 0 days ago
  • Low visibility — Only 8 GitHub stars
Code Pass
  • Code scan — Scanned 9 files during light audit, no dangerous patterns found
Permissions Pass
  • Permissions — No dangerous permissions requested

No AI report is available for this listing yet.

SUMMARY

CLI + Claude Code plugin that spends your Claude Code subscription 5-hour quota on real parallel work, stopping at a threshold. Metered against the real Anthropic usage endpoint.

README.md

dont-burn-it-all

CI
Release
Go Report Card
License: MIT

Spend your Claude Code subscription 5-hour quota on real work — many parallel headless
claude -p sessions — and stop at a threshold so you keep a reserve. Metered against the real
Anthropic usage endpoint, not a local estimate.

Install

# prebuilt binary via mise (no Go needed)
mise use -g "github:tbobm/dont-burn-it-all[exe=burn]@latest"

# or from source
go build -o burn .        # or: just build

Quick start

burn setup                                    # verify config (claude, token, endpoint, dirs)
burn --dry-run --goal test                    # print current 5h usage + planned jobs; run nothing
burn --goal "write tests for pkg/foo" --jobs 4 --target 80   # spend quota up to 80%
burn --watch --target 80                       # monitor only: notify when 5h usage hits 80%

Each --goal launch runs --jobs sessions to completion and stops — you start each launch.
Pick a --target above your current usage, or the launch refuses by design.

Flags

Flag Default Meaning
--target 25 Refuse/notify once 5-hour utilization reaches this %
--jobs 1 Parallel sessions per launch
--goal — Task each session works on (required to launch)
--model opus Model for sessions
--watch false Monitor mode: poll + notify, spawn nothing
--workdir temp scratch dir Where sessions run — a scratch dir, not a real repo
--store ~/.claude/burn/worker.jsonl JSONL log of sessions and readings
--max-turns 30 Max agent turns per session
--max-usd-guard 0 (off) Abort if reported cost exceeds this $
--dangerously-skip-permissions false Run sessions unattended (opt-in)
--i-know-this-bills-api false Override the billing-risk env refusal
--dry-run false Print state, spawn nothing

Safety

burn only spends subscription quota, never pay-per-token API. It refuses to run when a
billing-risk env var (ANTHROPIC_API_KEY, ANTHROPIC_AUTH_TOKEN, Bedrock/Vertex) is set, scrubs
them from every session, and runs a one-time preflight that proves a probe actually moves your
5-hour usage before launching real work. Sessions are interactive unless you pass
--dangerously-skip-permissions — only do that with a throwaway --workdir.

Example: prepare pending PR reviews

Spend idle quota drafting review comments in pending state (nothing submitted):

burn --jobs 1 --target 90 --dangerously-skip-permissions --goal '
Review PR https://github.com/OWNER/REPO/pull/123 (`gh pr diff 123`). Draft line-anchored
comments and create them as a PENDING review via `gh api .../pulls/123/reviews` with no
event — do NOT submit.'

Claude Code plugin

Ships a /burn slash command and a project skill (.claude/skills/burn/). Install it from any
Claude Code session — the repo is its own marketplace:

/plugin marketplace add tbobm/dont-burn-it-all
/plugin install dont-burn-it-all@dont-burn-it-all

Then say "burn quota" / "watch my usage" — the skill installs burn if missing, runs
burn setup, and drives it (optionally as a background sub-agent tracked with Monitor).

Development

mise install      # dev toolchain (Go + just)
just vet test     # go vet + go test
just dist         # cross-build release archives into dist/

Releases are automated with release-please
(Conventional Commits): a merged release PR tags a version, attaches per-platform binaries, and
publishes a ko image to ghcr.io/tbobm/dont-burn-it-all.

License

MIT — see LICENSE.

Reviews (0)

No results found