Prompt components

Drop-in micro-prompts you add to a user runner to enforce one specific execution behavior.

Prompt components are small drop-in add-ons you paste into a user runner (.user.txt) to add one specific execution behavior.

They are not standalone workflows.

Choose a component

Start with the one behavior you want to add to a prompt file. Components are for a single execution rule, not a full workflow.

Don’t confirm without evidence
Prevent automatic agreement with strong user assertions unless supported by evidence.
Open component
Read all provided artifacts first
Ensure the model reads the full user-provided text/files before answering.
Open component
Scan all artifacts exhaustively
Require exhaustive artifact coverage, explicit scan disclosure, and evidence pointers.
Open component
Browse the web and cite sources
Require retrieval from current/public sources when correctness depends on external facts.
Open component
Pause for analysis before answering
Add a short analysis pass before the final answer in higher-depth review or synthesis tasks.
Open component
Add a structured analysis pass
Add a deeper analysis/compliance layer for review, debugging, or comparison tasks.
Open component

Components catalog

Usage rules

Usage rules (show) Usage rules (hide)
  • Paste components into a user prompt file (`.user.txt`) unless explicitly documented otherwise.
  • Use 1–2 components max per run to reduce instruction collisions.
  • Components that require tools (for example browsing) should be used only in runtimes that support those tools.
  • Use components to add one behavior; do not treat them as full workflows.

Component details

anti-auto-agreement (user)
Prevents automatic agreement/confirmation of user assertions unless supported by evidence.
  • Use when: you need stance-neutral answers and want to avoid mirroring the user’s claim.
  • Avoid when: there is no user assertion or stance to evaluate.
Open component
deep-read (user)
Requires complete reading of user-provided artifacts before answering.
  • Use when: the task depends on provided text, files, or repository content.
  • Avoid when: there are no artifacts or inputs to read.
Open component
deep-scan (user)
Requires exhaustive artifact scanning, explicit coverage disclosure, and evidence pointers.
  • 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 component
deep-search (user)
Requires web retrieval and citations for external or up-to-date facts.
  • Use when: correctness depends on current/public sources and the runtime supports browsing.
  • Avoid when: you are in artifacts-only mode or the runtime has no browsing/tools.
Open component
analyze-before-answering (user)
Adds a short analysis pause before the final answer.
  • Use when: the task is non-trivial and you want a brief evidence/constraint/conflict pass before the final answer.
  • Avoid when: the task is trivial, latency-sensitive, or already uses a heavier analysis component.
Open component
deep-analyzed (user)
Adds a structured analysis pass with missing-input, contradiction, and compliance checks.
  • Use when: the task is review, debugging, comparison, or any higher-depth analysis.
  • Avoid when: the task is trivial and the extra overhead adds no value.
Open component