Vocabulary

Each term has one meaning. Use these terms in all framework files. Adapters map each term to a harness feature.

Term	Meaning	Do not use
harness	The software that runs the model, gives it tools, and controls the loop.	agent tool, IDE
model	The language model that the harness calls.	AI, LLM (in rules)
session	One run of a harness with one context. A fresh session has no context from another session.	chat, conversation
role	A named job in the workflow, with its own instructions, tools, and permissions. A human or an agent can fill it.	agent persona, sub-agent (in the core)
actor	A human or an agent that fills a role.	user, bot
driver	The actor that does the work in one stage: agent, human, pair, or checks.	owner (of a stage)
pair	A human and an agent in the same session. The human decides each change. Counts as one actor.	co-pilot, assist
mode	A preset of stage drivers: agent-centric, hybrid, or human-centric.	autonomy level
instruction file	The file that the harness loads into every session (repo level or user level).	memory, rules file
skill	A folder with a procedure that loads only when a task needs it.	command, prompt
hook	A script that the harness runs at a set event (for example before a file edit).	callback
check	A program that returns pass or fail with no model judgment.	test (in general), validation
eval	A repeatable test that measures how a component changes agent results.	benchmark
gate	A point where work stops until an approver (a human or a check) passes it.	approval step
stage	One step of the workflow model (for example Plan or Build).	phase
profile	The configuration of the workflow model for one type of work.	workflow type, mode
artifact	A file that a stage makes (for example a plan, a decision log, a test report).	output, document
plan artifact	A file that describes the approach before the build. A human can approve it.	spec (in the core)
decision log	A file where the agent records the decisions that it made without a human.	notes
Definition of Ready	The fields an input must have before work starts.	DoR
component	Any rule, skill, role, hook, check, or CI job that the generator makes.	asset
manifest	The file in the target repo that lists each component and its lifecycle data.	registry
adapter	The mapping from the neutral core to one harness or one tool.	plugin, driver
model profile	The idioms and defaults of one model family inside one harness adapter.	model config
idiom	One piece of vendor guidance for a model, with a source and an eval.	tip, trick
vendor	The company that releases a model or a harness.	provider
trigger	The event that starts a profile (a command, a label, a schedule, a CI event).	entry point
protected path	A path that a role must not change (for example tests for the implementer).	locked file
waiver	A recorded reason to skip a gate or a limit for one item.	exception, override
writing standard	The rules for human-facing text. The default is ASD-STE100 (STE).	style guide
human-facing text	Text whose main reader is a person (PR body, plan, report, reply).	docs (in general)
project dictionary	A repo file that gives each domain term one meaning.	glossary
persona	One type of person who uses the product: who they are, their job stories, their limits, and the evidence for them. It is not a role. See workflow-model.md.	user type, agent persona
job story	One thing that a persona must get done: "When I ..., I want to ..., so that ...". Its ID is <persona id>.J<n>.	user story (in the core)
requirement	One behavior that the product MUST have, with one or more WHEN and THEN scenarios. Its ID is <capability>.R<n>.	feature, story
spec	The current requirements of one capability. It is generated from the applied spec deltas. See workflow-model.md.	plan (for a spec), living document
spec delta	The requirements that one plan adds, changes, or removes. A human approves it with the plan.	patch, change set