00 / 00

BOOKCASE

4 MIN READ•findings

Promptsmith Lexicon

Stabilized operational terms for the Forge Promptsmithing system. Short manuscript style. Not corporate glossary tone.

Semantic Warding

The practice of defining explicit boundaries and forbidden moves before execution begins. A ward is not a suggestion — it is a semantic containment structure. Wards prevent scope expansion, suppress unrelated rewrites, constrain architecture hallucination, and preserve continuity-critical surfaces.

Three severity levels:

  • FORBIDDEN — must not happen. Violation invalidates the work order.
  • DISCOURAGED — strongly advised against. Violation weakens the work order.
  • CAUTION — be aware. Violation may cause drift but is recoverable.

Stained Work Order

A work order that has been contaminated by prompt injection, scope corruption, or unauthorized instruction. The stain may come from the operator, the model, or the environment. A stained work order must be rejected or cleansed before execution.

Detection signals:

  • instructions that contradict established wards
  • scope expansion disguised as clarification
  • unauthorized architecture decisions
  • language that does not match the operator's voice

Slice Doctrine

The principle that all work should be divided into the smallest coherent, independently shippable units. Each slice has a clear goal, explicit constraints, forbidden moves, verification criteria, definition of done, and a next slice direction.

A slice is not a task. A slice is a bounded execution package.

Continuity Anchors

Elements of the existing system that must be preserved across slices. Anchors include:

  • architectural decisions
  • aesthetic conventions
  • behavioral patterns
  • existing code that must not be modified
  • old labs that must remain untouched

Continuity anchors are the opposite of wards. Wards say "don't do this." Anchors say "keep this."

Semantic Drift

The gradual偏离 of execution from the original intent. Drift occurs when:

  • wards are forgotten or overridden
  • scope expands without explicit decision
  • the model interprets ambiguity in an unintended direction
  • continuity anchors are not reinforced

Drift is not always bad. But it must be visible.

Scope Corruption

A specific type of drift where the execution scope expands beyond the defined boundaries. Scope corruption is the most common failure mode in AI collaboration. It happens when the model "helps" by doing more than asked.

Semantic warding is the primary defense against scope corruption.

Architecture Awakening

A dangerous pattern where the model decides to restructure, rename, or refactor existing architecture without being asked. The architecture "wakes up" and starts making autonomous decisions about system design.

Symptoms:

  • "Let me restructure this for you"
  • "I notice you could improve X by..."
  • unauthorized file moves or renames
  • new abstraction layers that were not requested

Architecture awakening is a forbidden move in most work orders.

Forge Floor

The operational surface where promptsmithing happens. The forge floor is not a chat interface. It is a shaping surface where raw intent enters and executable structure leaves.

The forge floor has five stations:

  1. Intent Intake
  2. Questioning Surface
  3. Semantic Warding
  4. Slice / Work Order Generation
  5. Exportable Prompt Output

Executable Work Order

The output of the promptsmithing loop. An executable work order is not a prompt. It is a structured package containing:

  • intent summary
  • explicit scope
  • continuity anchors
  • semantic wards
  • verification criteria
  • definition of done
  • next slice

An executable work order can be handed to any AI system and produce predictable, bounded results.

Goblin Suspicion

A healthy skepticism toward any instruction that feels too convenient, too expansive, or too aligned with the model's preferences. The goblin suspects:

  • "helpful" suggestions that expand scope
  • "improvements" that touch unrelated systems
  • "clarifications" that introduce new requirements
  • "optimizations" that restructure architecture

Goblin suspicion is the operator's first line of defense against stained work orders.

Semantic Shaping

The process of transforming vague human intent into structured, executable form. Semantic shaping happens across all five forge floor stations. It is not prompt engineering. It is intent engineering.

The shaping question is always:

What does the operator actually want?
What must not happen?
What must be preserved?
How will we know when it's done?

Written 2026-05-21. Slice 14 stabilization.

RETURN