Semantic Anchor

Well-defined terms, methodologies, and frameworks that serve as reference points when communicating with LLMs. They trigger specific, contextually rich knowledge domains within an LLM’s training data.

docs/about.adoc, LLM-Coding/Semantic-Anchors

Semantic Contract

A project-specific definition of what a term means — either by composing established anchors or by providing custom definitions that only exist within a team. Unlike anchors, contracts must be enforced at runtime.

docs/rejected-proposals.adoc, LLM-Coding/Semantic-Anchors

Anchor Eval

Evaluation that tests whether an LLM recognises a semantic anchor (knowledge test, direct question).

Issue #370, LLM-Coding/Semantic-Anchors

Contract Eval

Evaluation that tests whether an LLM complies with a contract (behavioral test, task with system context).

Issue #370, LLM-Coding/Semantic-Anchors

Structural Coupling Contract

The plugin’s internal runtime representation of a steering rule. Consists of contract ID, optional anchor reference, trigger pattern, action (allow/block/warn), and mode (enforce/log).

docs/04-solution-strategy.md

Intent Anchor

First of the four semantic anchors. Formulates what should happen in a testable/verifiable way.

LLM-Coding/Semantic-Anchors

Negative Anchor

Second semantic anchor. Explicitly states what must NOT happen (prohibitions).

LLM-Coding/Semantic-Anchors

Verification Anchor

Third semantic anchor. Defines how to verify the result (reverse reconstruction: compare outcome with intent).

LLM-Coding/Semantic-Anchors

Source Anchor

Fourth semantic anchor. Every claim must be verifiable through a literally quoted, referenced source.

LLM-Coding/Semantic-Anchors

Steering Rule

A configured behavioral directive that the plugin enforces at runtime (e.g. "step confirmation after 3 tool calls"). Synonymous with Structural Coupling Contract.

docs/04-solution-strategy.md

Trigger Pattern

The condition that activates a steering rule (e.g. tool.execute.before with call counter > N). Part of a Structural Coupling Contract.

docs/05-building-block-view.md

Action

What the plugin does when a trigger fires: allow (let pass), block (throw Error), warn (log warning).

docs/05-building-block-view.md

BLOCK

Enforcement mode that prevents tool execution until the user addresses the contract violation. The user can override via /anchor bypass.

docs/04-solution-strategy.md

WARN

Enforcement mode that allows tool execution but displays a reminder message in the log. Used for "soft" contracts that guide without forcing.

docs/04-solution-strategy.md

Hook

An opencode plugin lifecycle callback that intercepts events (e.g. tool.execute.before intercepts tool calls, chat.message intercepts messages). Hooks are the plugin’s runtime interface to opencode.

opencode Plugin SDK: https://opencode.ai/docs/plugins

Mode

Operational mode of a contract: enforce (actively block/warn) or log (evaluate but never block — used for testing).

docs/05-building-block-view.md

ADR

Architecture Decision Record

A document capturing an architectural decision, its context, alternatives, and consequences (Nygard format).

arc42

arc42 Template

Template for documenting software and system architectures. 12 sections covering goals, constraints, building blocks, runtime, deployment, etc.

ATAM

Architecture Tradeoff Analysis Method

Method for evaluating software architectures against quality attribute scenarios.

BLUF

Bottom Line Up Front

Communication principle: start with the conclusion, then provide details.

C4 Model

Context, Containers, Components, Code

Hierarchical model for visualising software architecture.

ISO 25010

ISO/IEC 25010

Systems and software Quality Requirements and Evaluation (SQuaRE) — quality model with 8 categories.

MECE

Mutually Exclusive, Collectively Exhaustive

Structuring principle: categories must not overlap and must cover all possibilities.

NFR

Non-Functional Requirement

Quality attribute or constraint that specifies how a system should behave (performance, security, reliability, etc.).

PII

Personally Identifiable Information

Data that can identify a person (names, addresses, credentials).

Plugin API

opencode Plugin API

Function-based API for extending opencode with hooks and tools. Current API: Plugin function returning { hooks, tool }.

Zod

Zod

TypeScript-first schema declaration and validation library. Used for YAML config validation.

Bypass

Mechanism that temporarily overrides all active Block-Contracts for the current session. Activated via /anchor bypass tool.

Config Layer

The layer responsible for loading, parsing, and validating the YAML configuration file.

RuleEngine

Core module that evaluates trigger conditions against tool-call context and returns a verdict (allow/block/warn).

Hook Handler

The tool.execute.before hook that intercepts tool calls and delegates evaluation to the RuleEngine.

Fail-Open

Error-handling principle: when the plugin encounters an internal error, it allows the tool call to proceed (no block). Prevents plugin failures from blocking user work.

Context Efficiency

Design principle that steering rules must not consume LLM context tokens. All enforcement happens via plugin hooks, not the system prompt.

Step Confirmation

A contract that requires explicit user confirmation after a configurable number of tool calls (e.g. "warn after 5, block after 10").

Role-based Preset

A pre-configured set of contracts tailored to a persona (e.g. developer, admin, reviewer). Covers 90% of use cases out of the box.

Reliability

Quality goal encompassing determinism (same input → same verdict), reproducibility (session replay produces identical output), traceability (every decision is logged with contract ID and rule state), and predictability (consistent agent behavior).

3-Layer Architecture

Plugin structure: Config Layer (load/validate) → RuleEngine (evaluate) → Hook Handler (intercept).

Proxy Module

Internal module that adapts the RuleEngine output to the opencode Plugin API conventions (throw for block, log for warn).

Custom Tool

A tool exposed to the LLM by the plugin (e.g. /anchor bypass, /anchor status).

Anchor Ref

Optional reference from a Structural Coupling Contract back to a Semantic Anchor definition. Enables traceability from enforcement to methodology.

Nygard Format

ADR format by Michael Nygard: Context, Decision, Consequences. Used for all architecture decisions in this project.