Runner Low Burn v2

Cost-control operating profile for review-gated autonomous coding runners. Stops avoidable token burn from idle supervisor polling, duplicate reviews, and full-context replays while keeping proof gates and review intact.

Source

superada.ai/skills/runner-low-burn-v2

Entrypoint

SKILL.md

Installability

Ready for agent install

Source spec

superada:runner-low-burn-v2

This page is designed to be the thing you send to another agent. It tells the agent where the source of truth lives, whether install is real today, and what command or manual path to follow.

Canonical source

https://superada.ai/skills/runner-low-burn-v2/

curl -sSf https://superada.ai/install/runner-low-burn-v2 | sh

Install the skill described on this page from superada:runner-low-burn-v2. Review the source first, then run: curl -sSf https://superada.ai/install/runner-low-burn-v2 | sh

1. 1
   Step 1

   Install the skill from the SuperAda installer.

2. 2
   Step 2

   Load it on top of Runner when burn, not scope, is the problem.

3. 3
   Step 3

   Apply the polling, envelope, and review budgets before swapping models.

4. 4
   Step 4

   Escalate with evidence-first receipts when a stop-loss trips.

• Skill contract
  https://superada.ai/skills/runner-low-burn-v2/SKILL.md

  Low-burn operating rules, budgets, and escalation contract.

• Context exhaustion recovery
  https://superada.ai/skills/runner-low-burn-v2/references/context-exhaustion-recovery.md

  Recovery ladder when worker context runs out mid-issue.

• Credential blocker closeout
  https://superada.ai/skills/runner-low-burn-v2/references/credential-blocker-closeout.md

  Closing out credential-blocked issues without burning tokens.

• Native goal low-burn launch
  https://superada.ai/skills/runner-low-burn-v2/references/native-goal-low-burn-launch.md

  Launching native goal-mode runs under low-burn budgets.

• Curl installer
  https://superada.ai/install/runner-low-burn-v2

  Installer for local agent runtimes.

• Does not remove review gates or proof requirements.
• Not a substitute for well-shaped work manifests.
• Stop-loss rules can halt a queue; restart requires operator review.