executing-plans
executing-plans
Use when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available
escalate-or-holdDecide whether an issue gets escalated now, held with a re-check time, or closed — as a typed decision with probabilities rather than a judgment call in a thread. Use when asked should I escalate this, is this a P1, do we page someone, hold or escalate, or when an on-call, support lead or PM is staring at an issue and unsure. Produces the issue state in a f…
Independent GitHub snapshot · Automated classification · Original documentation governs
Most escalation mistakes are timing mistakes: too late because nobody wanted to be the one who paged, or too early because the reporter shouted. This skill fixes the shape of the decision — six state fields, three defined options, a probability for each — so the call is made on impact and trend rather than on volume.
Before cs-escalation-brief (which writes the brief once you have decided to escalate) and alongside escalation-tree (which says to whom). This one says whether, and when.
escalate / hold / close with probabilities and confidenceAsk for these if not provided:
| Option | It applies when |
|---|---|
escalate | Impact is growing, or a commitment / SLA / safety line is crossed, and the current owner cannot fix it in time — bring in the next level now |
hold | Impact is bounded, the owner has a credible plan and a next update time — hold and re-check at that time |
close | Resolved, a duplicate, or impact was misjudged — close with a note to the reporter |
Thresholds. Act when confidence ≥ 0.6 and the winning option's probability ≥ 0.7. When escalate and hold are close, hold with a hard re-check time and tell the reporter the time — a close call resolved by a clock beats a close call resolved by a mood.
Trend beats severity. A bounded sev-2 that is getting worse is closer to escalation than a static sev-1 with a tested fix landing in an hour.
Getting probabilities. Any calibrated decision model that returns a probability per defined option serves this contract; the repo ships an adapter (node integrations/jev/decide.mjs escalate-or-hold --state state.json, contract in integrations/jev/decisions/). Without one, the assistant estimates and labels the estimate.
State
| Field | Value |
|---|---|
| impact | … |
| owner | … |
| trend | … |
| prior_escalations | … |
| asks | … |
Decision
| Option | Probability |
|---|---|
| escalate | 0.xx |
| hold | 0.xx |
| close | 0.xx |
Confidence: 0.xx · Source: [model / estimate]
Verdict: [act: option] or [hold — re-check at HH:MM; escalate before that if: …] Reporter line: "[one sentence: what is happening and when they will hear next]"
escalate