Escalate stalled BUILD or QA processes by upgrading to a stronger model.
npx skills add https://github.com/jackson2787/torque-coding --skill escalate-jackson2787
Please help me install this Agent Skill. Skill: escalate Source: https://github.com/jackson2787/torque-coding/tree/main/skills/state-machine/escalate Command: npx skills add https://github.com/jackson2787/torque-coding --skill escalate-jackson2787
Escalation handles stalled BUILD or QA cycles by upgrading to a stronger model while preserving the memory-bank context so progress can continue without losing task state.
Trigger escalation automatically when a stall is detected, or manually switch to a stronger model and re-enter the project to resume BUILD or QA.
High-intent search queries and answers about installing and using this skill.
Escalating stalled build or QA processes involves upgrading to a stronger model while preserving memory-bank context. It spawns a subagent on a higher rung using a configured ladder and passes an escalation-brief to resume the task.
You should escalate stalled builds or QA cycles when they stall due to repeated errors or time limits. The escalation mechanism detects these stalls and records progress before upgrading the model to continue execution.
Configuring the model ladder requires defining rung limits in the .memory-bank-v2/machine/limits.md file. This configuration dictates which stronger model the subagent will spawn with when an escalation is triggered.
If subagent execution is unavailable, the fallback path gracefully instructs you to manually switch to a stronger model and re-enter the project to resume the build or QA task, ensuring no data loss.
Escalation attempts are tracked by writing an escalation-brief.md file at a specified location and recording each attempt in the current-task/build-log.md, providing a full audit trail across ladder steps.
Yes, the resume flow returns control to the BUILD or QA process after an escalation completes. This allows the upgraded model to re-check the task state and finish the stalled work.
jackson2787Generate a deterministic plan_context.md from an approved plan.md.
emcconnellVerifies completed stories' acceptance criteria, design consistency, and prepares completion reports.
ssmleoAuthor append-only source-authority memory separating trusted and untrusted evidence.
KiingoDiagnose stalled prospect stall types and generate low-pressure re-entry messages.
