Full catalog
Search 中文
More
AI VISTA / SITE NAVIGATION

Courses and topics open directly from the main navigation.

Open the full content catalog

Courses

4
  1. 01Build your first testable AI workflow5 lessons
  2. 02Make AI answers reliable enough to use14 lessons
  3. 03Design answers that can show their evidence11 lessons
  4. 04Automate with permission, recovery, and cost controls37 lessons
View all courses →

Topics

9
  1. 01Task framing4 guides
  2. 02Context and prompts6 guides
  3. 03Evaluation7 guides
  4. 04Grounded answers11 guides
  5. 05Agent architecture12 guides
  6. 06Agent safety5 guides
  7. 07Local model deployment7 guides
  8. 08Model operations7 guides
  9. 09Field cases4 guides
View all topics →

Context and prompts

What must the model see and return for the next step to work?

Map instructions and sources, preserve decisions as conversations grow, define acceptable content and a consumable output format, then test untrusted content.

AI VISTA / DECISION01 / 03
WORKING DECISIONSeparate missing facts, unclear instructions, and unusable output shapes; each needs a different fix.
01INPUT
02CHECK
03BOUNDARY
RETURN TO THE REAL TASK✓

DECISION BRANCH

Answer the decision this topic actually owns.

  1. ASK FIRST↓

    Does the model actually receive the facts, rules, and source precedence needed for each output field?

  2. IF↓

    The material is complete but behavior varies: write an acceptance contract and compare changes on the same cases.

  3. OTHERWISE✓

    Supply or mark missing material before extending the prompt to hide an input gap.

WORKED CASE

A contract summary keeps missing renewal terms in the annex.

The model receives a master agreement, two amendments, and an internal review policy, but file versions and precedence are unstated. The output looks complete while a renewal term comes from a superseded annex.

WORKING ARTIFACTA context-flow map connecting source, selection, precedence, and output fields.
01WEAK SHORTCUT

Keep lengthening the prompt with ‘be accurate’ and ‘do not omit anything,’ then place every file into one undifferentiated context.

02BETTER JUDGMENT

Map the context first: current agreement, valid annexes, internal policy, user question, and output fields. Declare precedence, missing states, and the source location expected for each conclusion.

03ACCEPTANCE

Test 12 agreements containing version conflicts. Amount, date, renewal, and termination fields point to valid clauses; uncertainty is explicitly held for review.

ENTRY & EXIT SIGNALS

Know when to enter—and when the decision is good enough to leave.

A topic is not an endless knowledge directory. Entry signals tell you whether the problem belongs at this layer. Exit signals decide whether to continue instead of substituting time spent reading for work completed.

AI VISTA / DECISION STATUSContext and prompts
01ENTER HERE
  • The same task omits different content across inputs
  • Source conflicts produce unexplained choices
  • A longer prompt does not reduce consequential failures
02LEAVE WHEN
  • Every output field maps to required context
  • Source precedence, missing states, and format are declared
  • Failure is no longer summarized as a bad prompt

6 PRACTICES

Build the judgment, make the artifact, then review it with real cases.

01 Map the context before rewriting the promptSeparate instructions, source material, examples, history, tools, and output rules to see what the model actually receives. Context and prompts 02 How many tokens should one LLM call reserve for input and output?Count the next complete request, reserve room for tool results and all generated tokens, then check model input, output, and context limits before sending it. Context and prompts 03 When should you compact a long AI conversation?Set a compaction checkpoint from measured context use, preserve decisions and unresolved actions, then test whether the next turn still follows the facts. Context and prompts 04 Write the acceptance contract before the promptDefine required fields, evidence, tone, uncertainty, and rejection conditions before optimizing prompt wording. Context and prompts 05 Choose JSON Schema, Markdown, or plain text for AI outputMatch an AI response format to its downstream consumer, then route refusals, truncation, parser errors, schema errors, and factual errors separately. Context and prompts 06 Test prompt injection in documents and tool outputsBuild a regression set for untrusted pages, files, retrieval snippets, and tool results; check intended behavior and permission gates before release. Context and prompts

BOUNDARY

What this topic page will not do

This page makes inputs and instructions visible. It cannot supply facts that were never provided or replace source governance.

01Do not hide missing material with a longer prompt02Do not treat citation-shaped text as evidence03Do not build RAG before source precedence and version rules exist
When every output field maps to an input, rule, or explicit missing state, turn the context map into evaluation cases.