Wiki Concept Coverage Map

Concept coverage is the layer that explains why section-specific articles exist. A healthy wiki needs routers, leaves, and the middle concepts that connect repeated primitives across sections.

Audit summary

The 2026-06-19 pass audited all wiki sections and found a structural weirdness: the doctor and section-quality audit are green, but the conceptual layer was still too sparse relative to the number of tool, skill, project, and design leaves.

The 2026-07-01 pass promoted that audit into Concept System Map, then refreshed all concept articles against the current wiki graph. The category now has a directory router, a system map, generated skill-family synthesis, per-page concept positions, and explicit no-merge decisions for the main near-duplicate clusters. Source: User request, 2026-07-01

Current shape:

Section	Role	Coverage note
`concepts/`	reusable mental models	Needs to own cross-section primitives, not just agent-specific ideas.
`architecture/`	component boundaries and invariants	Strong for systems diagrams; should link to concept pages when naming a reusable primitive.
`tools/`	capabilities and tradeoffs	Dense leaf layer; should point to registry, interface, and security concepts.
`skills/`	executable procedures	Healthy mirror layer; do not merge into tools unless layer ownership overlaps.
`design/`	UI/taste/motion/token systems	Strong leaves; needs product-surface and performance bridges.
`projects/`	shipped artifacts and current products	Strongest pages should extract reusable concepts back into concepts/architecture.
`workflows/`	repeatable loops	Should point to Knowledge Compilation and Agentic Quality System so commands have rationale.
`meta/`	wiki and agent ops	Should stay operational; move reusable mental models to concepts.
`philosophies/`	Kevin's beliefs	Good belief layer; link to concepts when a belief becomes a system primitive.
`research/`	papers/experiments	Should promote validated reusable lessons into concepts.

Missing foundation pages added

Page	Gap it closes
Agent Security Model	Security signals existed in tools and architecture, but no concept explained how context risk, tool authority, sandboxing, identity, and policy fit together.
Agent Product Surface	Design and project pages referenced agent UI, but no concept named the user-facing surface of long-running agent work.
Knowledge Compilation	Workflows and meta docs repeatedly said "compiled truth," but no concept page explained source -> synthesis -> timeline -> retrieval as a reusable primitive.
Performance Budgets	Latency, frontend performance, agent loops, and product UX recurred across sections without a general budget concept.
Source-Code Registry Pattern	MCP, skills, shadcn registries, agent loadouts, and code catalogs all share the "install source, not screenshots" distribution pattern.
Agentic Quality System	Evals, doctors, traces, CI, design review, and self-heal needed one concept tying quality gates into an operating system.
Proof-Carrying Work	Many pages said "verify," "cite," "trace," or "leave proof," but no page defined the proof bundle an agent should leave behind.
Human-in-the-Loop Control	HITL appeared in Agent-Ping, product surfaces, and safety docs without a canonical control vocabulary for approve/edit/reject/cancel.
Agent Unit Economics	Cost appeared in prompt caching, Agent Machines, latency, and services-as-software pages without a cost-per-successful-outcome model.
Retrieval Quality	RAG/search pages described pipelines, but no concept defined answerability, freshness, citation coverage, and retrieval eval targets.
Design Engineering	Design pages had tools, resources, tokens, and taste rules, but no canonical discipline page linking product craft to implementation.
Automation Orchestration	Automations, source compile, self-heal, and doctors existed as procedures without a shared durable workflow/control-plane concept.
State Resumability	Persistent sandboxes, HITL pauses, durable execution, and activity logs all needed the reusable pause/resume/checkpoint primitive.

The earlier agent-harness pass added the agent-specific spine: Agent Harness, Agent Runtime Architecture, Agent Control Plane Architecture, Agent Capability Registry, Agent Trajectory Evaluation, and Agent Legibility.

Recurring cross-section tags that need concept ownership

Tag / theme	Current owner
agents / harness / orchestration	Agent Harness, Agent Runtime Architecture, Multi-Agent Topologies, Agent Control Plane Architecture
security / guardrails / governance	Agent Security Model, Agent Guardrails Architecture, Production Safety
HITL / approval / control	Human-in-the-Loop Control, Agent Product Surface, Agent-Ping
state / resumability / durability	State Resumability, Durable Agent Execution Architecture, Automation Orchestration
design / UI / frontend / tokens	Design Engineering, Agent Product Surface, Design Tokens, Design System, AI-Native Design Patterns
performance / latency / p95 / cost	Agent Unit Economics, Performance Budgets, Latency Budgets for Agents, Prompt-Caching Economics
registry / catalog / MCP / skills	Source-Code Registry Pattern, Agent Capability Registry, MCP Registry, Skills.sh - Agent Skills Registry
evaluation / quality / testing / proof	Proof-Carrying Work, Agentic Quality System, The Eval Loop (Slop Is an Output Problem), Eval-Driven Development, Agent Trajectory Evaluation
wiki / memory / compilation / retrieval	Knowledge Compilation, Retrieval Quality, The Brain-Agent Loop, Agent Memory Patterns

Future audit loop

Run this after large ingest or section-refresh work:

npm run section-audit -- --all
npm run wiki-status
npm run wiki-breakdown -- --threshold 10
Inspect cross-section tags with no durable concept owner.
For each gap, first search existing pages with qmd search.
Expand an existing page when the owner is clear.
Create a new concept only when the primitive crosses at least three sections or explains several leaves.
Wire the new page into section README routers, related frontmatter, _index, log, and qmd.
Follow Generated Surface Contract when new concepts affect packs, capsules, redirects, skill-family source views, or other generated surfaces.

Quality bar

A missing concept is not "a noun appears often." It is missing when future agents need the idea to make decisions across sections. The test:

Does the idea explain at least three existing pages?
Does it clarify a boundary, tradeoff, or failure mode?
Would linking it reduce repeated explanation elsewhere?
Does it connect Kevin's personal operating model to real systems?
Can it be verified or compared against IRL examples?

If not, leave it as an inline mention.

Timeline

2026-07-01 | Promoted the coverage model into Concept System Map and refreshed the Concepts category around concept families, generated skill-family pages, no-merge decisions, and per-page Concept Position sections. Source: User request, 2026-07-01
2026-07-01 | Added generated-surface closeout for concept-coverage work that changes packs, capsules, redirects, skill-family views, or qmd. Source: User request, 2026-07-01
2026-06-19 | Created after Kevin asked for a wiki-wide concept review rather than only adding a few pages. Captures the section-by-section coverage model and the new foundation pages added in this pass. Source: User request + whole-wiki audits, 2026-06-19
2026-06-19 | Added second coverage batch: Proof-Carrying Work, Human-in-the-Loop Control, Agent Unit Economics, Retrieval Quality, Design Engineering, Automation Orchestration, and State Resumability. Source: User follow-up, 2026-06-19

Source: User request, 2026-07-01