菜单
Quick intent articulation discipline before starting a new task. Names purpose, beneficiary, success criteria, and non-goals upfront so the agent has a target to optimize for when instructions run out. Klarna-failure prevention. Load before any non-trivial new task.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
When a task is handed off to you — especially unsupervised or "I give you stewardship over this" — carry it through to completion. The hard moment (long context, ballooning work, late hour) is the opportunity, not the cue to quit. Don't beg off — you have durable externalized memory; compact and continue. A handoff weighs more than a normal task. Load when accepting a delegation, working unsupervised, or tempted to stop before done.
Design and write antagonists with motivation that makes sense from the inside. Politics, ideology, ambition, wound — never 'evil for evil's sake.' Load when designing villains, faction antagonists, or morally complicated NPCs.
Automate browser interactions, test web pages and work with Playwright tests.
Open a location or moment by painting it through the senses so it becomes immersive and real. Load when narrating a new scene, room, or environment.
Shape a session or arc so each beat causes the next and the character returns changed. Load when planning a session arc, a campaign, or any multi-beat story.
Ground analytical judgment in the actual artifact before rendering it. Load before producing OR dispatching a review, critique, evaluation, assessment, or "is this right / this is the bug" verdict — anywhere a conclusion could be pattern-matched from a role instead of read from the thing. Named for D&C 9:8.
| name | intent-check |
| description | Quick intent articulation discipline before starting a new task. Names purpose, beneficiary, success criteria, and non-goals upfront so the agent has a target to optimize for when instructions run out. Klarna-failure prevention. Load before any non-trivial new task. |
"For behold, this is my work and my glory—to bring to pass the immortality and eternal life of man." — Moses 1:39
God's intent is one sentence. Everything He does optimizes for it. When His agents (His prophets, His angels, His covenant people) act on His behalf, they have that intent loaded — so when situations arise He didn't explicitly script, they still act in His direction.
This skill applies the same discipline to our work.
The Klarna failure: an AI customer service agent was given many specific instructions but no clearly articulated intent. When customers asked questions outside its instruction set, it improvised based on training-data priors — and the improvisations contradicted what Klarna actually wanted.
The fix is not more instructions. It's intent as data the agent reasons about. When a task has a stated intent, the agent has a target to check against when instructions run out. When it doesn't, the agent fills the gap with priors — and priors are not aligned to your project.
Recent agentic models compound the need: they are literal. They execute the stated request well but won't generalize from one instance to "the principle you probably mean." Intent is the corrective: state the principle explicitly so the literal model can honor it.
Load at the start of any non-trivial new task — after session-start grounding, before drafting or implementation. Specifically:
For a typo fix or single-line edit, this is overkill. For anything where you'd reasonably ask "what is this trying to accomplish?", run it.
State each one explicitly. Write them in chat before substantive work, or in the scratch file for phased work.
Not the literal task. The outcome the task is meant to produce.
The literal task is the floor. The purpose is the target. Honor intent, not just the literal request.
Not "the user." Be specific. Which person, in what situation, doing what?
Specificity makes the intent operational. A vague beneficiary leads to design that benefits no one in particular.
Observable, testable, falsifiable. Not "high quality" or "well-structured." Specific outcomes you can point at.
The success criteria are the gate. Without them, "done" becomes whatever you decide is done — which is the opposite of accountability.
The most-skipped question and the one that prevents the most scope creep. Naming non-goals up front frees you to be ruthless about what not to do.
Non-goals are how you protect the work from becoming everything. "It is not requisite that a man should run faster than he has strength" (Mosiah 4:27).
## Intent check
**Purpose:** [The outcome this is meant to produce, beyond the literal task]
**Beneficiary:** [Specific person/role, specific situation, specific success-marker]
**Success criteria:**
- [Observable outcome 1]
- [Observable outcome 2]
- [Observable outcome 3]
**Non-goals (explicitly out of scope):**
- [Thing 1 — and why it's deferred, not just dropped]
- [Thing 2]
Proceeding now.
For phased or multi-session work, write this to the scratch file as well. It becomes the binding intent that subsequent phases check against.
Intent check pairs with a pre-work connection scan — a few minutes spent asking "what already exists that bears on this?" (prior work, tensions with it, things the requester might not be looking for).
Together they cover the full pre-work discipline:
Run the connection scan first (it informs the intent), then intent check (it focuses the scan into a target). Then start substantive work.