Prompt Components for AI Workflow Controls

Reusable prompt components for adding evidence boundaries, verification checks, deep search, artifact reading, confidence scoring, and no-fabrication controls to AI workflows.

Prompt components are single-purpose instruction fragments you add to a prompt, project instruction, or reusable workflow to enforce one specific execution behavior.

Add them when you need one specific behavior such as artifact reading, broader public-source search, or a deeper analysis pass. Use 1–2 components per run to reduce instruction collisions, and use tool-dependent components only in runtimes that support those tools.

Browse components by behavior

Pick one behavior, open the component page, or jump to the detailed usage notes below.

Don’t confirm without evidence
Prevent automatic agreement with unsupported user assertions.
Open file Details
Read all provided artifacts first
Ensure the model reads the full user-provided text or files before answering.
Open file Details
Scan all artifacts exhaustively
Require exhaustive artifact coverage, explicit scan disclosure, and evidence pointers.
Open file Details
Search public sources exhaustively
Require broader public-source retrieval when correctness depends on external or current facts.
Open page Details
Cited authoritative sources
Require public-world factual claims to use authoritative sources with stable locators.
Open page Details
Terminology consistency control
Define key terms from supplied material and prevent terminology drift across the output.
Open page Details
Analyze before answering
Add a brief analysis checkpoint before the final answer.
Open page Details
Run deep analysis
Separate grounded observations, verified claims, and NOT VERIFIED items before concluding.
Open page Details
Block fabricated or simulated output
Prevent invented facts, fabricated evidence, simulated runs, guessed results, and unsupported claims.
Open page Details
Use an objective technical register
Keep responses professional, objective, technical, structured, and free from persuasive or narrative framing.
Open page Details
Add a confidence score
Append an evidence-based confidence line to standard responses.
Open page Details
Return copy/paste-ready structure
Force stable headings, bullets, tables, checklists, code blocks, or exact replacement blocks when useful.
Open page Details

Component details

Component
Don’t confirm without evidence
do-not-confirm-without-evidence.md
  • Use when: you need stance-neutral answers and want unsupported user claims treated as hypotheses until evidence supports them.
  • Avoid when: there is no user assertion, conclusion, interpretation, or stance to evaluate.
Open file
Component
Read all provided artifacts first
read-all-provided-artifacts-first
  • Use when: the task depends on provided text, files, logs, or repository content.
  • Avoid when: there are no artifacts or inputs to read.
Open file
Component
Scan all artifacts exhaustively
scan-all-artifacts-exhaustively
  • Use when: you have multiple artifacts and need high confidence that relevant content was not missed.
  • Avoid when: there are no artifacts to scan.
Open file
Component
Search public sources exhaustively
deep-search.md
  • Use when: correctness depends on external, current, niche, or public sources.
  • Placement: add it to the user or workflow instruction layer for runs that require broader public-source retrieval.
Open page
Component
Cited authoritative sources
cited-authoritative-sources.md
  • Use when: public-world factual claims require authoritative sources and stable locators.
  • Placement: add it to the system, developer, project, or workflow instruction layer when public factual claims need cited support.
Open page
Component
Terminology consistency control
terminology-consistency-control.md
  • Use when: key terms need stable definitions and consistent use across a draft, answer, or review.
  • Placement: add it to the system, developer, project, or workflow instruction layer when terminology drift must be prevented.
Open page
Component
Analyze before answering
analyze-before-answering
  • Use when: the task is non-trivial and you need to identify evidence, constraints, conflicts, or uncertainty before answering.
  • Avoid when: the task is trivial, latency-sensitive, or already uses a heavier analysis component.
Open page
Component
Run deep analysis
deep-analysis.md
  • Use when: the task requires objective technical analysis, claim separation, and explicit NOT VERIFIED items.
  • Placement: add it to the user or workflow instruction layer for runs that require a stronger analysis pass.
Open page
Component
Block fabricated or simulated output
no-simulation-no-fabrication.md
  • Use when: the task involves facts, logs, citations, code behavior, test results, source analysis, or any claim that must be grounded.
  • Avoid when: the task is explicitly fictional, hypothetical, or creative and the fictional scope is clearly labeled.
Open page
Component
Use an objective technical register
objective-technical-register.md
  • Use when: the output must stay professional, technical, scoped, and free from persuasive or narrative framing.
  • Avoid when: the task intentionally requires marketing, storytelling, community-facing copy, or emotional tone.
Open page
Component
Add a confidence score
confidence-score.md
  • Use when: the answer should expose evidence strength, uncertainty, and remaining limitations.
  • Placement: add it to the system, developer, project, or reusable instruction layer.
Open page
Component
Return copy/paste-ready structure
structured-copy-paste-ready-output.md
  • Use when: the answer must be reused directly in implementation steps, documentation, QA plans, review checklists, or remediation work.
  • Avoid when: the user explicitly asks for prose, narrative writing, or a non-structured format.
Open page