Personalized Research Glossary & Learning Flywheel

A scoped agreement defining the question, audience, decision, evidence standard, exclusions, output, and validation criteria.

A repeating cycle of framing, evidence gathering, synthesis, publication, observed use, and new questions.

A time-bounded execution with its own inputs, sources, findings, and changes.

A structured register connecting claims to sources, quality, dates, confidence, and unresolved conflicts.

A presentation-independent representation of concepts, relationships, claims, decisions, and evidence.

An addressable unit such as a concept, claim, decision, example, control, procedure, or source.

A small self-contained unit designed for linking, reuse, and targeted revision.

A statement asserted to be true and therefore requiring support proportional to its consequence.

A directly noticed pattern or result without an asserted causal explanation.

A conclusion derived from observations and evidence rather than directly stated by a source.

A proposed course of action grounded in evidence, constraints, alternatives, and consequences.

A forward-looking statement whose uncertainty and disconfirming indicators should be explicit.

A condition treated as true for analysis even though it has not been fully verified.

A consistent method for labeling unknowns, confidence, conflicts, and what would change a conclusion.

A named state such as observed, supported, contested, inferred, unverified, or superseded.

Checking a consequential claim against multiple independent forms or sources of evidence.

An original specification, official document, repository, dataset, changelog, or direct record.

Evidence from widely used code, reference implementations, or production case studies.

A credible case that contradicts or narrows a general claim.

The property that a claim can be tested and potentially disproven.

A meaningful baseline, alternative, or prior state used to interpret a claim.

How current a source, claim, implementation, or recommendation remains.

The exact version, commit, date, or edition of a source used by the research.

A focused comparison between the current knowledge state and a newer research run.

A deliberate attempt to disprove findings, expose blind spots, and test failure modes.

Recording unresolved questions, discarded paths, exceptions, and partial findings.

Advancing toward execution, then returning upstream when evidence exposes a framing problem.

Distinguishing a visible defect from the upstream model, rule, or process that produced it.

Ask only when the answer could materially change routing, evidence, architecture, risk, or validation.

An observation, problem, opportunity, incident, or strategic change that may deserve structured attention.

A compact orientation stating context, importance, constraints, and the immediate decision or investigation.

Request for Discussion: exploration of a problem space, options, and unknowns before a formal proposal.

Request for Comments: a proposal seeking review on a defined technical or organizational direction.

Architecture Decision Document: the artifact that captures the selected direction and rationale.

Architecture Decision Record: the durable historical record of a decision, context, consequences, and status.

The current description of architecture, components, interfaces, flows, constraints, and operations.

A broader implementation-oriented explanation combining requirements, architecture, trade-offs, and rollout.

A precise description of required behavior, interfaces, data, constraints, and acceptance conditions.

An agreement about interface, behavior, policy, ownership, or quality.

Tests, datasets, metrics, thresholds, and review procedures for judging behavior.

Sequenced workstreams, dependencies, owners, milestones, migration, and rollback.

Operational Readiness Review: a gate proving the system can be operated safely and reliably.

A record of what shipped, with versions, dependencies, approvals, tests, and artifacts.

Evidence that the deployed system still matches approved contracts, policies, and architecture.

A post-delivery assessment of outcomes, surprises, failures, and required changes.

A documented, time-bounded exception with owner, rationale, risk, and expiry.

The explicit replacement of an older decision, artifact, or rule by a newer one.

A managed state indicating something remains available but should no longer be adopted.

A structured unit containing the decision, alternatives, evidence, rationale, owner, status, and consequences.

A decision model defining Driver, Approver, Contributors, and Informed parties.

Explicit authority to recommend, approve, implement, operate, or accept risk.

A decision that can be changed with limited cost, blast radius, or disruption.

A decision with high switching cost, lock-in, regulatory impact, or broad blast radius.

The authoritative location for a class of artifact or state.

Rules represented in a versioned, testable, deployable form rather than only prose.

A required proof or approval before progressing to a later lifecycle state.

A status such as Draft, Proposed, Approved, Implementing, Validated, Deprecated, or Superseded.

A lightweight path for low-risk, familiar, reversible work.

A formal path for consequential, cross-cutting, regulated, or hard-to-reverse work.

A bounded path for learning with hypotheses, constraints, and exit criteria.

The deliberate design of audience, knowledge, evidence, narrative, representation, interaction, and validation.

Structuring information, alternatives, evidence, consequences, and next actions for a specific decision.

The arrangement of decisions, dependencies, authority, timing, and evidence across a process.

The organization of concepts, relationships, levels of detail, and retrieval paths.

A representation of each audience’s context, expertise, incentives, constraints, questions, and actions.

A projection of the same canonical knowledge tailored to a role’s decisions and depth needs.

The opening context that says where the reader is, why it matters, and what the artifact enables.

Presenting the central finding or recommendation before detailed support when evidence is sufficient.

A purposeful sequence from context and tension through evidence, choices, consequences, and action.

The reason a section exists: orient, explain, compare, prove, warn, decide, act, or preserve evidence.

Revealing detail in stages while keeping essential meaning and actions visible.

Increasing conceptual difficulty in a deliberate learning sequence.

Providing a stable system-level view before focused component or scenario views.

Signals that help a reader predict where a link, section, or control will lead.

A stable link target for a concept, claim, section, or decision.

A primary linear path through a complex knowledge graph.

Exploration through relationships among concepts rather than only hierarchical menus.

Using meaningful concept selections and relationships to guide authoring or exploration.

The immediately related concepts around a selected knowledge object.

Limiting the number and depth of simultaneous exploration choices.

Navigation whose position, labels, and hierarchy remain predictable across views.

Starting paths such as Author, Approver, Reviewer, Operator, Executive, Architect, or Engineer.

The least favorable realistic combination of viewport, attention, connectivity, capability, authorization, and interruption.

Changing hierarchy, sequence, and interaction for smaller contexts rather than simply shrinking.

A visible way to understand failure, undo, retry, or resume.

A structured way to disclose, propagate, and archive corrections across artifacts.

A compact model for determining purpose, audience, decision, evidence, constraints, and representation.

A table connecting claims or requirements to tests, owners, evidence, and release criteria.

One knowledge base rendered into coordinated executive, architecture, engineering, and operational views.

An effect produced by consistency, evidence, structure, and deliberate restraint rather than visual heaviness.

HTML elements chosen for meaning, structure, accessibility, and machine interpretation.

A portable artifact with styles, scripts, metadata, and content in one file.

A client-side application running from static files without a dedicated application server.

Starting with usable semantic content and adding interaction without making it a prerequisite.

Using details and summary for expandable content.

A native HTML modal or nonmodal interaction surface.

A browser-native mechanism for transient overlays such as menus and panels.

Declarative HTML commands connecting controls to dialogs, popovers, and actions.

An inert DOM fragment stored for later cloning or transformation.

A named reusable HTML element with encapsulated behavior.

Custom elements, templates, shadow DOM, and slots used as platform-native components.

Shadow DOM expressed in generated HTML rather than created only by JavaScript.

Content remaining in the regular document tree, often preferable for research, search, and print.

A Web Components insertion point that separates content from component layout.

An abstract syntax tree representing document sections, blocks, lists, figures, and claims.

A limited grammar of approved content and layout primitives.

A generated presentation derived from a canonical knowledge model.

A named value for color, spacing, typography, shape, motion, or elevation.

A runtime variable used for reusable values and theme behavior.

An explicit ordering mechanism for groups of CSS rules.

A style query based on the size or capabilities of a containing component.

A controlled baseline that normalizes inconsistent browser defaults.

Styles adapting the artifact for paper or PDF output.

Browser storage for small preferences or prototype data; not secure or synchronized.

Structured title, authorship, dates, version, provenance, and content semantics.

JSON-based linked data embedded in HTML.

A shared vocabulary for describing entities and content on the web.

Metadata used by social and messaging platforms to create link previews.

A packaging model for research objects, data, software, and provenance.

A file-packaging convention for reliable transfer and preservation of digital content.

An immutable published instance of an artifact at a known point in time.

Preserving successive research events and deriving the current state from them.

Structured JSON or similar data included alongside rendered content.

A publication designed to remain useful without network access.

An underlying system of shared edges, rails, gutters, and baselines established before styling.

Consistent repetition of spacing, alignment, typography, and grouping patterns.

A bounded scale and pattern for margins, padding, gaps, and line spacing.

A stable alignment line used by headings, body text, icons, numbers, and controls.

A visual correction used when mathematical centering appears misaligned.

A fixed area for a number, badge, icon, or status beside a content stack.

Align a leading number, badge, or icon with the heading rather than the full multi-line content block.

Center the visible icon shape, not merely its font box or SVG viewport.

Draw a connector with a shaft and arrowhead in dedicated whitespace.

Reserve space between process components so connectors remain centered and clear.

An arrow character that is useful inline but unreliable as structural geometry.

Related components sharing anatomy, spacing, states, and semantics.

A bounded set of structural arrangements selected by communication purpose.

A reading-oriented composition using hierarchy, pacing, and narrative sections.

A monitoring-oriented composition optimized for state, comparison, filters, and drill-down.

A scan-oriented composition with stable navigation, definitions, examples, and cross-links.

Changing the opening composition to match the artifact’s purpose.

Size, position, spacing, contrast, and grouping used to signal importance and sequence.

A limited hierarchy of text sizes, weights, and line heights.

A controlled line length supporting comfortable reading.

A variable typeface used as a case study for embedding, subsetting, licensing, and offline packaging.

A font file containing axes such as weight, width, or optical size.

Removing unused glyphs or ranges to reduce font size.

A CSS font list using fonts already installed on the device.

The design of symbols intended to communicate across cultures and languages.

Pairing text with an icon when meaning or consequence could be ambiguous.

A symbol-only control requiring a familiar icon, accessible name, and clear state.

One header control cycling light, dark, and system modes.

A restrained glow or gradient used to create focus without carrying meaning.

How crowded or calm an interface feels beyond raw element count.

A widely used accessibility conformance target covering perceivability, operability, understanding, and robustness.

A 4.5:1 minimum contrast target for typical body text under WCAG AA.

A 3:1 minimum contrast target for sufficiently large or bold text under WCAG AA.

A 3:1 target for meaningful UI boundaries, controls, and graphical objects.

Testing every component and state rather than sampling only page-level text.

A user-agent mode replacing authored colors with a system high-contrast palette.

A clearly perceivable indicator showing which interactive element has keyboard focus.

The ability to reach and operate functionality without a pointing device.

The programmatically determinable label announced for an interactive element.

A semantic page region such as header, navigation, main, aside, or footer.

Logical nesting of headings representing document structure.

Roles, states, and properties used when native HTML semantics are insufficient.

Choosing a built-in HTML element before implementing a custom control.

A user preference indicating nonessential animation should be minimized.

Keeping content, diagrams, controls, and code within their intended bounds.

Testing the smallest viewport for horizontal scroll, clipping, and collapsed hierarchy.

Actually exercising buttons, dialogs, filters, exports, and state changes.

Checking measured alignment, dimensions, overlap, and containment.

Comparing rendered output with a baseline to detect unintended changes.

A view of which features work, degrade, or require fallbacks in target browsers.

A way to describe when a browser feature is broadly available.

Cross-browser work focused on consistent web platform capabilities.

Testing pagination, expanded content, links, contrast, and metadata in print.

A checklist of default, hover, focus, active, disabled, selected, warning, success, and error.

Consistent sizing, alignment, hit target, and icon placement of controls.

A repeatable set of artifact-specific checks performed before delivery.

A workflow retrieving the relevant artifact profile and prior battle scars for preflight and release.

A generalized rule derived from a recurring failure or costly correction.

Checks performed before implementation to prevent known failure patterns.

Maintaining core meaning and operation when advanced capabilities are unavailable.

Model Context Protocol: a client-server protocol for exposing tools, resources, prompts, and related capabilities to AI applications.

The application-side component that discovers and invokes MCP server capabilities.

A process exposing capabilities while enforcing its own authorization and execution rules.

The user-facing AI application coordinating model interaction, clients, permissions, and context.

An invocable operation described by purpose and an input schema.

Addressable contextual content exposed by a server.

A reusable prompt template or workflow exposed by a server.

The communication mechanism used between client and server, locally or remotely.

An MCP server running in the same developer environment as the client.

An MCP server reached across a network boundary with enterprise authentication and tenancy controls.

Reveal only capabilities relevant to the current task, expanding as needed.

A searchable catalog of tool metadata, ownership, risk, scopes, and endpoints.

Retrieve capability descriptions in bounded pages rather than loading an entire catalog.

Select the smallest relevant capability set based on intent, identity, policy, and context.

Machine-readable metadata describing what a tool or agent can do and how to invoke it.

An emerging pattern for richer interactive experiences associated with MCP capabilities.

Emerging approaches for exposing web-page capabilities to agents through structured interfaces.

Agent-to-Agent protocol concepts for discovery, task exchange, coordination, and results.

A machine-readable description of an agent’s identity, skills, endpoints, and expectations.

Defined states for an agent task such as submitted, working, input-required, completed, failed, or canceled.

A durable result object produced during or after an agent task.

An acronym used by multiple agent communication or commerce initiatives; issuer and specification must be resolved.

Answer Engine Optimization: structuring content so answer systems can understand, retrieve, and cite it.

An emerging use of HTTP payment-required semantics and associated payment flows for machine-accessed services.

A commercial or state-changing action initiated through an agent for a user or organization.

An additive capability layered onto a base specification.

An exchange determining supported versions, features, and constraints.

The degree to which a tool, server, agent, or workflow can move across hosts, vendors, and runtimes.

Code intended to demonstrate or validate a specification.

Tests verifying whether an implementation follows required protocol behavior.

Divergence between specification versions, vendor extensions, and deployed implementations.

A recurring check of versions, proposals, changelogs, and adoption.

Establishing the identity of a user, workload, client, or service.

Determining whether a principal may perform a specific action on a resource.

A framework for delegated authorization using access tokens and defined roles.

An identity layer on OAuth providing authenticated user information.

Microsoft’s cloud identity and access platform for workforce identity, applications, policy, and token issuance.

A credential presented to a resource server to authorize API access.

A token describing an authenticated session for the client, not a general API credential.

A credential used by a client to obtain new access tokens.

A compact signed token format containing claims.

A named assertion in a token such as subject, tenant, role, scope, or audience.

A delegated permission string describing allowed API operations.

The intended recipient API or resource server for a token.

The trusted authorization server that created and signed a token.

Proof Key for Code Exchange, binding an authorization request to its token exchange.

A browser-mediated flow exchanging a short-lived code for tokens.

A workload obtains a token using its own identity rather than a user’s.

A constrained client or CLI completes sign-in through a separate browser.

On-Behalf-Of: a delegated exchange obtaining a downstream token in the signed-in user’s context.

Exchanging one security token for another with a different audience or delegation context.

Permission exercised by an application in the context of a signed-in user.

Permission granted directly to a workload independent of a user.

A client capable of securely holding credentials, such as a server-side application.

A browser or installed client unable to reliably protect a secret.

A cloud-managed workload identity obtaining tokens without embedded credentials.

The identity assigned to a service, function, container, or automation.

A trust mapping that exchanges an external workload identity for a cloud token.

A trusted service acquiring, exchanging, caching, and supplying tokens to authorized execution components.

Protected storage for tokens keyed by user, tenant, scope, and audience.

Using managed or federated identities instead of long-lived application secrets.

Tokens, cookies, and secrets never enter prompts, model context, or model-visible results.

Rechecking authorization immediately before a consequential action.

Identity policy evaluating user, device, location, risk, and application.

Reacting to important identity or policy changes before token expiry.

A token cryptographically bound to a client key rather than usable by any bearer.

A proof-of-possession mechanism using signed HTTP proofs.

Mutual TLS, where client and server authenticate with certificates.

A privileged service is tricked into misusing its authority for another party.

The point where a user or administrator authorizes access or an action.

The target scenario where an MCP service accesses Jira under the user’s delegated identity.

A controlled entry point for model invocation, policy, routing, observability, and cost attribution.

Separating application behavior from a single model or provider API.

A stable interface mapping platform requests to provider-specific APIs.

A normalized description of model features such as modality, tools, streaming, and structured output.

A governed catalog of approved models, capabilities, versions, regions, cost, and risk.

Switching to an alternate provider, model, or degraded mode when the preferred path fails.

Selecting models, tools, data, or controls based on identity, use case, risk, cost, geography, and rights.

Classifying AI uses by consequence, exposure, data sensitivity, and autonomy.

A lower-exposure path for employee or developer use with bounded data and review.

A higher-exposure path affecting customers, public content, transactions, or brand.

Data and tool access evaluated using the authenticated user’s or workload’s identity.

Retrieval enforcing content rights, territory, time, person, contract, or audience constraints.

A unified path where policy, identity, audit, and approvals are consistently applied.

A supported implementation path with templates, controls, examples, and operational defaults.

A production-oriented example demonstrating the approved architecture and controls.

A deterministic or model-assisted control constraining inputs, outputs, tools, or actions.

Cleaning, bounding, or labeling untrusted prompt inputs before model use.

Systematic measurement of model and application behavior using representative tasks and criteria.

Evaluation against fixed datasets or scenarios outside live traffic.

Evaluation using live or shadow traffic, outcomes, and runtime signals.

Tracing prompts, model choices, latency, cost, errors, tool use, and evaluated outcomes.

Assigning model, retrieval, tool, and infrastructure cost to teams, products, users, or workflows.

Reusing provider-side or application-side processing for repeated prompt prefixes or context.

Retrieval-Augmented Generation: retrieve external knowledge and supply it to a model for grounded generation.

Using entities and graph relationships to support broader or multi-hop retrieval and synthesis.

A managed or custom system for ingesting, indexing, retrieving, and citing organizational knowledge.

A system indexing embeddings for similarity search.

A numeric representation used to compare semantic similarity.

Combining lexical, semantic, metadata, or graph signals.

Reordering retrieved candidates using a stronger relevance model or scoring step.

A trace from generated output to the supporting source passage or object.

The trace of how data, content, models, prompts, tools, and transformations produced an output.

A durable architectural rule guiding choices across products and teams.

Shared AI capabilities should support multiple product experiences.

Teams add capabilities within governed boundaries without central bottlenecks.

An end-to-end route from use-case intake through identity, models, controls, evaluation, and operations.

Coordinating model calls, tools, memory, state, approvals, and task progression.

A precise definition of tool purpose, inputs, outputs, errors, permissions, and side effects.

Model or tool output constrained to a defined schema.

A vocabulary for validating JSON structure and constraints.

Repeating an operation with the same key does not create duplicate effects.

A state change outside the immediate computation, such as creating, deleting, sending, or purchasing.

A workflow where a person reviews, supplies information, or approves before continuation.

A deterministic checkpoint before a high-impact or irreversible action.

Content that may contain misleading, malicious, or conflicting instructions.

Instructions embedded in untrusted content attempting to redirect model behavior.

Manipulative content designed to cause unsafe or unintended tool invocation.

Ordered precedence of system, developer, user, and untrusted-content instructions.

An isolated execution environment with bounded resources and access.

An explicit set of tools or operations permitted for a context.

The deterministic component evaluating policy before allowing an action.

A durable record of identity, request, decision, action, result, and context.

A linked record of steps across model calls, retrieval, tools, services, and decisions.

A bounded interaction context containing identity, state, and conversation continuity.

Information retained across steps or sessions for continuity or personalization.

Temporary context needed for the current task.

Persisted knowledge intended to survive across sessions.

An explicit model of allowed states and transitions.

A persisted intermediate state from which work can resume.

A defined maximum duration for an operation.

Reattempting a failed operation under defined conditions.

Increasing delay between retries.

A control temporarily stopping calls to an unhealthy dependency.

A bound on request frequency or volume.

Slowing producers when consumers or dependencies are saturated.

A queue for messages that repeatedly fail processing.

An operation reversing or offsetting a completed side effect when atomic rollback is unavailable.

Evidence that a principal performed or approved an action and cannot plausibly deny it.

The non-LLM layer owning identity, policy, schemas, routing constraints, and release rules.

A managed function runtime scaling per invocation.

A packaged process with its runtime, dependencies, and filesystem view.

The choice between request-oriented functions and longer-lived packaged services.

A managed HTTP entry point with routing, throttling, authentication integration, and observability.

AWS container orchestration for services and tasks.

AWS-managed Kubernetes.

Virtual machines with full operating-system control.

AWS content delivery and edge distribution service.

Lambda functions triggered at CloudFront edge locations.

A globally distributed serverless runtime based on web platform APIs.

A framework for stateful AI agents on Workers.

A uniquely addressable stateful compute instance on Cloudflare.

A globally distributed key-value store optimized for read-heavy data.

Cloudflare’s managed SQLite-compatible database.

Cloudflare object storage with S3-compatible semantics.

Cloudflare’s vector database service.

Cloudflare’s model inference platform integrated with Workers.

Managed messaging for asynchronous processing.

Durable orchestration for multi-step processes.

Cloudflare capability for isolated code execution environments.

The Cloudflare CLI for developing, deploying, and managing Workers resources.

Akamai’s edge-compute runtime for request and content logic.

A programmable HTTP accelerator and reverse proxy.

Startup latency when a runtime creates or initializes an execution environment.

Reducing active compute to none when idle.

The number of operations processed simultaneously.

Sending partial results over an open connection as they become available.

A one-way HTTP stream from server to client.

A bidirectional persistent connection.

Restricting services and dependencies to controlled network paths.

Policy governing outbound network destinations and data movement.

Deploying service capacity and data across regions.

The requirement that data be stored or processed in defined geographies.

A platform combining Postgres, authentication, storage, realtime, and serverless functions.

Authentication services for users, sessions, providers, and email-based sign-in.

Postgres policies filtering rows based on the current identity and operation.

A passwordless sign-in link delivered by email.

A provider limit on how frequently authentication emails can be sent.

Using an application without an authenticated account.

The product experience before authentication, which should explain value before emphasizing login.

The stable application frame shown after login, including identity and account controls.

A user-controlled workflow removing the account and associated user data.

A user-controlled export of stored information in a portable form.

A disclosure of collected data, purposes, retention, sharing, rights, and contact practices.

Rules governing acceptable use, availability, liability, and user obligations.

Atlassian’s managed platform for building apps across Jira and Confluence.

Declarative configuration of modules, permissions, functions, resources, and egress.

Forge’s Atlassian-native component and rendering model.

A Forge mode hosting a custom frontend within Atlassian products.

Server-side code executed in Atlassian’s managed runtime.

Managed app storage associated with the Atlassian installation.

Asynchronous processing for work exceeding interactive execution constraints.

Capabilities for updating app experiences as events occur.

A pattern connecting Forge apps to external services.

Atlassian-provided model access within the Forge ecosystem.

An Atlassian agent integrated with enterprise work context.

A callable operation available to a Rovo agent.

Atlassian’s graph of work entities and relationships.

A Forge API mode acting with the current user’s permissions.

A Forge API mode acting with the app installation’s permissions.

The missing identity, privacy, limits, recovery, accessibility, operations, and deletion work.

The ability to infer internal system behavior from emitted signals.

A timestamped event record.

A numerical time-series measurement.

A causal path of spans across a distributed operation.

An identifier propagated across logs and services for one request or workflow.

A vendor-neutral standard and toolkit for telemetry instrumentation and export.

An observability platform used for logs, traces, dashboards, alerts, and workflows.

An open-source observability platform built around OpenTelemetry.

Service Level Indicator: a measured signal representing service behavior.

Service Level Objective: a target for an SLI over a defined period.

The allowed amount of unreliability implied by an SLO.

How quickly an error budget is being consumed.

Operational instructions for diagnosis, mitigation, recovery, and escalation.

An unplanned event degrading service, safety, or business outcomes.

A classification of incident consequence and urgency.

A review of incident timeline, contributing factors, response, and corrective actions.

A contributing condition whose removal would materially reduce recurrence.

Reduced responsiveness caused by excessive or low-value alerts.

Grouping repeated events into one actionable incident signal.

Maintaining a reduced but useful service when dependencies fail.

Returning to a previously known-good version or state.

A runtime control enabling or disabling behavior independently of deployment.

A rapid control for disabling a risky capability or action path.

Deliberately introducing failures to verify resilience and recovery.

Testing system behavior under expected and extreme demand.

A meaningful change between expected and actual data, model, policy, or runtime behavior.

The named team accountable for operating, monitoring, and improving a system.

Tests, dashboards, runbooks, controls, and approvals proving operational preparedness.

Using dependency and request lineage to narrow where a failure entered a flow.

A structured definition of content types, fields, relationships, and constraints.

Content represented as typed fields and relationships rather than only rich text.

The organization, labeling, navigation, and retrieval structure of information.

A controlled classification of concepts into categories and subcategories.

A formal representation of entity types, relationships, and constraints.

A graph of entities and relationships.

A uniquely identifiable person, organization, system, concept, artifact, or place.

Determining when different names or records refer to the same entity.

Dividing source content into retrievable units.

A retrieval constraint based on rights, date, geography, author, tenant, or type.

Retrieval based on meaning similarity rather than exact terms.

Retrieval based on words, phrases, and term statistics.

A model or function scoring retrieved candidates for final relevance.

The exact passage, lines, or object supporting a claim.

Making content understandable and citable by answer engines.

Search Engine Optimization for discoverability in traditional search results.

Generative Engine Optimization, an emerging label for visibility in generative answers.

Structured data embedded in a page using a recognized vocabulary.

A declared preferred URL for substantially similar content.

A machine-readable list of site URLs and optional update metadata.

A site-level file communicating crawler access preferences.

An emerging convention proposing curated guidance for model-oriented content access.

How clearly a passage states a claim, context, source, and stable anchor.

Content that directly resolves a concrete question before expanding into detail.

Creation, update, review, and source-version dates.

Attributes describing who may access, use, transform, or distribute content.

Structured information describing origin, authorship, transformation, and evidence.

A catalog of sources, specifications, repositories, changelogs, and monitoring locations.

A qualitative measure of how strongly repeated evidence indicates a trend or priority.

A chronological view of signals, changes, and momentum.

A reusable package of instructions, methods, examples, and resources guiding an agent for a domain task.

The primary instruction file in an Agent Skills package.

Agent Skills-compliant release metadata stored under a supported metadata object.

Major, minor, and patch increments according to compatibility and change scope.

A curated history of meaningful changes across versions.

The upstream commit, version, or date from which a skill was derived.

A lightweight decision layer selecting the relevant domain skill or workflow.

A diagnostic embedded only where ambiguity could change a consequential decision.

Proceed with stated assumptions unless a missing answer could materially change the result.

The one main reasoning or communication technique selected for the task.

A secondary method used only to support the primary technique.

A bounded experiment or question used to test an uncertain direction.

A reversible extension layered onto an existing skill instead of replacing the base.

Keeping authorization, provenance, rights, release, and approval controls outside adaptable AI reasoning.

A structured provider-neutral representation of intent, context, constraints, workflow, and output.

A transformation from PromptSpec into provider- or runtime-specific instructions.

A prompt compiler targeting one provider’s conventions and capabilities.

A compact routing pass proposing ranked workflows from a request.

The built artifact for route selection, workflow configuration, saved sets, and provider compilation.

A configurable diagnostic level such as Smart, Thorough, Off, or Review-only.

A reusable bundle of prompt and workflow configuration.

A record of how a compiled prompt, model, tools, and evaluation produced an outcome.

A reusable set of tests and criteria associated with a prompt or workflow.

A browsable system of addressable knowledge objects, relationships, preferences, and provenance.

An explicit editable collection of durable user or team preferences.

User-specific information used to adapt responses, workflows, and artifacts.

A category such as current-task state, session context, durable preference, or institutional knowledge.

A design choice among reusable methodology, autonomous workflow, and packaged conversational experience.

A catalog of reasoning, research, communication, and validation techniques.

Difficult, ambiguous, or failure-oriented cases used to test a skill.

Visible version, update date, and source revision included in a skill or artifact.

The study and design of interactive systems around human capabilities, tasks, and contexts.

The discipline of structuring and labeling information for findability and comprehension.

The visual representation of abstract data and relationships.

Making complex technical information usable for specific audiences and tasks.

Communicating evidence, uncertainty, and findings to varied audiences.

Communicating hazards, uncertainty, actions, and trade-offs.

The study of judgment, choice, uncertainty, and decision processes.

The study of attention, memory, perception, learning, and reasoning.

Designing systems around human performance, limits, error, and operating context.

Designing end-to-end services across people, processes, policies, and touchpoints.

Reasoning about components, relationships, feedback, delays, and emergent behavior.

Creating, organizing, sharing, preserving, and reusing organizational knowledge.

Organizing, preserving, retrieving, and governing information.

Systematic design of learning outcomes, sequence, practice, and assessment.

How visual form shapes interpretation, credibility, and persuasion.

The study of signs, symbols, and meaning.

The study of effective and ethical persuasion for an audience and context.

Using data, visualization, narrative, and sourcing to explain events or systems.

Planning, creation, governance, and lifecycle management of content.

Research methods used to understand people, contexts, problems, and opportunities.

Observing representative users completing tasks with a system.

Systematic design and testing for diverse capabilities and assistive technologies.

Reaching agreement across different interests, constraints, and authority.

The arrangement of roles, decision rights, interfaces, and incentives.

How people understand, supervise, trust, and recover from AI behavior.

A role framing complex problems, defining reference architectures, aligning stakeholders, and guiding implementation pathways.

The mission area for foundational AI capabilities shared across engineering teams.

Treating shared capabilities as a product with users, roadmap, adoption, support, reliability, and outcomes.

Tools, documentation, examples, support, and workflows helping engineers adopt a platform.

A staged route from discovery through trial, integration, validation, and production use.

A reusable pattern showing components, flows, controls, and trade-offs.

A durable rule guiding choices across systems.

Shared understanding of requirements, architecture, constraints, decisions, and consequences.

Making technical, security, operational, and organizational risks explicit to decision-makers.

Advancing sound decisions through evidence, facilitation, trust, and shared incentives.

Raising material concerns with evidence, alternatives, and a path forward.

Communicating complex judgment clearly, calmly, and proportionally to senior decision-makers.

Reasoning from user outcomes, adoption, feedback, lifecycle, and value.

Making ownership, dependencies, and unresolved gaps visible without silently taking over adjacent roles.

Help close missing work while explicitly recording the ownership or capacity gap.

A definition of mandate, responsibilities, exclusions, authority, interfaces, and success measures.

Responsibility for priorities, capacity, approvers, trade-offs, and risk acceptance.

Responsibility for implementation, code, tests, integration, delivery, and operations.

Responsibility for technical direction, patterns, decisions, alignment, and risk visibility.

The arrangement of roles, processes, governance, funding, and feedback delivering a capability.

An indicator showing whether an initiative achieved its intended outcome.

The primary long-term result used to align related work.

Concrete work required to create immediate value or evidence.

The durable target state and principles guiding future evolution.

Support for multiple teams, tenants, regions, policies, and operational demands.

A consistent developer interface across approved model providers.

CLI, IDE, and agent tools operating under the developer’s enterprise identity.

Controls delivered as reusable defaults, APIs, templates, and evidence rather than only approvals.

A recurring synthesis of significant changes, context, implications, risks, and next watch points.

A broad view derived from a maintained registry rather than one announcement.

A maintained list of official changelogs, specifications, repositories, releases, and trusted reporting.

A structured list of organizations, products, model families, and update channels.

Detecting, verifying, classifying, and interpreting changes over time.

A sustained directional pattern supported by multiple signals over time.

A discrete observation that may contribute to a trend.

Classifying a signal by domain, maturity, consequence, audience, and time horizon.

A visual encoding of accumulated evidence or momentum.

A visual sequence of signals and inflection points.

A branching visual map of concepts and relationships.

A hierarchical visual representation of related concepts or phrases.

Showing a compact source summary with expandable detail.

Merging repeated coverage of the same underlying event or claim.

Separating sponsored content from editorial or technical evidence.

Distinguishing code that can be operated directly from managed vendor capability.

Changes and signals relevant to immediate awareness.

Patterns visible across several days.

Structural movement and adoption patterns.

Long-cycle shifts in standards, architecture, regulation, and market structure.

The verified delta from the prior known state.

The technical, business, security, or organizational consequence of a change.

Indicators, proposals, releases, or adoption evidence that would change the interpretation.

A domain, vendor, protocol, geography, or source type missing from the registry.

The reduction in relevance or predictive value of an older signal.

A qualitative judgment based on source diversity, duration, and credible alternatives.

Evidence moving against the dominant interpretation.

Choosing a SPA, deck, dashboard, or template before clarifying the decision and audience.

Reusing the same hero, cards, and dashboard structure across unrelated content.

Including material because it was found rather than because it serves orientation, proof, decision, or action.

Forcing a long clarification process on every task.

Applying many frameworks simultaneously because each seems useful.

Importing an external taxonomy wholesale into every skill.

Replacing an entire SKILL.md to integrate a new idea.

Treating the published page as the only research record.

Letting a model freely emit final HTML and CSS without a constrained structure.

Delivering after visual inspection without exercising interactions.

A generic descendant CSS rule unintentionally styling icon or layout wrappers.

Placing arrow characters between cards so they overlap or sit on borders.

Centering a number or icon against a multi-line title-and-body stack.

Styling components independently without shared edges and rails.

Allowing same-type cards or boxes to use different internal spacing.

Multiple sticky headers or controls overlap because offsets were not coordinated.

Keeping desktop hierarchy and simply reducing dimensions.

Displaying every citation or source detail in the primary reading flow.

Placing many identical act-now blocks throughout an artifact.

Allocating excessive width to short number or rank columns.

Checking a few page colors but not component states and themes.

Using visual separator elements inconsistently or without semantic function.

Relying on a temporary file link without preserving or regenerating the artifact.

Completing analysis but failing to produce the promised file.

Allowing an operational quota to appear only as an error.

Passing OAuth tokens, cookies, or credentials through prompts or model-visible data.

Assuming an authenticated conversation authorizes all later tool actions.

Loading a very large tool list into model context.

Adding login without redesigning data, states, export, and deletion.

Treating an acronym such as ACP as having one universal meaning.

Publishing time-sensitive guidance without an update mechanism.

Ending design at components and flows without ownership, runbooks, SLOs, or recovery.

Implementing governance only as centralized approvals.

Separating citations and support from the statements they justify.

Every consequential operation should preserve or explicitly transform the identity and authority under which it executes.

Every consequential claim should retain a path to its source, state, date, and uncertainty.

Permissions, territory, time, and usage constraints should remain attached through retrieval, generation, and delivery.

Exploration, proposal, decision, design, implementation, readiness, runtime, and learning are distinct linked states.

The canonical knowledge model should exist independently of HTML, deck, dashboard, or prompt.

Identity, authorization, rights, provenance, validation, and release controls should not depend solely on model judgment.

Stage detail without deleting evidence, alternatives, exceptions, or recovery.

Deployed behavior must be tested and measured; a correct diagram is insufficient.

Failures, retries, undo, rollback, correction, and resumption should be designed into systems and artifacts.

The compliant path should include reusable tools, defaults, evidence, and support.

Design and validate for constrained realistic conditions, not only the ideal desktop or network.

Agents and teams act independently inside explicit capabilities, policies, budgets, and recovery mechanisms.

Practical portability depends on auth, extensions, semantics, conformance, deployment, and operations.

Reference and adopted implementations reveal missing lifecycle, security, and operational details.

A local correction should become a compact preflight and release rule.

Useful personalization needs preferences, relevance, provenance, correction, and deletion.

A useful brief explains verified change, why it matters, and what to watch next.

Layout, hierarchy, diagrams, spacing, and interaction communicate relationships, confidence, consequence, and action.