跳至正文
OpenAgentSkill
浏览已开放 · 目录记录日期 2026-11-21 · 自动更新和账户服务暂时关闭。
其他技能

sitrep

Use when the user brings several updates at once — "quick catch-up", a list of things that happened, news touching the plan, people, time and money together. Sorts each update to the skill that writes it, shows the routing for a yes, then works through each one in the next turn. Writes only the log itself.

skyf0xx · 已有静态检查记录

安装与使用

运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。

npx skills add skyf0xx/gambit --skill sitrep

技能说明


name: sitrep description: Use when the user brings several updates at once — "quick catch-up", a list of things that happened, news touching the plan, people, time and money together. Sorts each update to the skill that writes it, shows the routing for a yes, then works through each one in the next turn. Writes only the log itself. display: plain-card writes: log requires: goal phase: run next: plan, stakeholders, capacity, decide, strategy, recon

Skill: sitrep

Trigger: The user brings several updates at once. "Council said no, Sam dropped out, I'm down to £2k." A catch-up after a few days away. Any message where two or more separate things changed.

Purpose: Take the batch in one pass. Split it into facts, send each to the skill that owns it with route_updates, show the routing for a yes, then work through each one the next turn with a light update, not a rerun.

This skill writes only the log. It owns no section of the goal.


Voice & Tone

Quick and tidy, like a good chief of staff taking notes. No lecture on what each update means; the owning skill does that if it needs to. Short lines, plain words. If one fact is bigger than the rest, say so.


Execution Sequence

1. Split the Message

Read the goal and the user's message. Break the message into discrete facts, one change each. "Sam dropped out and I'm short on cash" is two facts.

Drop anything that changes nothing: venting, a repeat of what the goal already says.

2. Pick the Owner for Each Fact

The fact is...Owner
A move done, blocked or dropped; a new date on a moveplan (Mode 0, status flip)
Someone joining, leaving or changing role in what the user is doingintake writes people; stakeholders moves a person across and keeps people[].status current
Someone with a say changed their stancestakeholders
Hours, money, energy or runway changedcapacity
A new fork the user has to choose atdecide
A question the user had is answeredrecon (answer mode)
A conversation happened and has an outcomenegotiate (debrief mode)
A test or a bet has a resultexperiment or forecast
The facts together move what matters moststrategy

Route only to skills in the table. route_updates takes six items at most and works only while sitrep is active.

Things no key holds. A preference, a limit, a fact about the user's life that an update carries with it: call remember right now, in this turn, before routing. A move they refuse goes in as rejected.

Too big to route. If a fact breaks something the goal stands on (the focus no longer holds, the deadline moved by a month, the goal itself changed), do not route it. A light update would paper over it. Name the full skill as the next step instead.

3. Show the Routing as One Confirm

Call route_updates({items: [{skill, update}]}), each update one short plain sentence. Then reply once with kind: confirm:

  • say: what you'll update, short. "I'll drop the permit move, move Sam off the team and log £2k as your runway."
  • bottomLine: the yes you need. If a fact was held back as too big, say so here.
  • options: Yes / Not quite.

Write nothing to the goal this turn apart from remember.

4. Work Through the Routed Updates (next turn)

On a yes, each routed skill is cleared to write in the same turn it loads. Load them one at a time. For each, make the light update that skill's own quick mode would make, and no more:

  • plan: Mode 0 status flip or date change on the matched move. No replan.
  • capacity: change the numbers the user gave. No fresh interview.
  • stakeholders, intake: move or edit the one person.
  • decide: record the new fork as an open decision.
  • recon: set the question answered, with the answer.
  • negotiate: debrief mode on the matching prep entry.
  • experiment, forecast: record the result on the matching entry.

On "Not quite", ask what is wrong in one line, fix the routing and confirm again. Skip any item the user turned down; remember it as rejected if it was a move.

When one update changes what another skill's section rests on, say so in the reply. Do not chase it this turn.

5. Log Only What Was Decided

By now a routed skill is active, not sitrep, so any log entry comes from the routed skill that made the change, as its own quick mode would. Facts the owning keys hold are not repeated. If nothing was decided, write no entry.

6. Close with the Single Next Move

Call finish_skill after the last routed skill. Then one reply: bottomLine is the one next move, usually the item the updates exposed or the full skill a held-back fact needs. options carry the alternatives.

Next: [one move, usually the item the updates exposed]

Or:
  - A fact was too big to route -> the skill that owns it (often strategy)
  - A plan rests on something that moved -> plan
  - Get the whole picture back in plain words -> brief