Quicky Wiki boundary

Type: agentic-system-analyses/types/agentic-system-boundary.md

Boundary and evidence

Quicky Wiki is in scope as a memory/knowledge/context-engineering system. Its intended use is to turn source documents into a retained, confidence-scored knowledge base that people and agent clients can query. This is not merely a library API or an MCP tool: the repository ships the callers that initialize local state, ingest documents, maintain claims, render pages, and later reload and consume them. The boundary is whole-system, covering these shipped paths, their embedded prompts and their user-facing entry points. This classification rests on SRC-1, package.json, src/cli.ts, src/lib.ts, src/cli/context.ts, src/compiler/ingest.ts, src/graph/store.ts, src/graph/query.ts and src/mcp/server.ts.

The declared purpose is explicit (SRC-1, README.md):

Turn any collection of documents into a living, confidence-scored knowledge base — powered by LLMs. --- README.md

The implementation reads source content, extracts claims through a model, compares them with retained knowledge and resolves the result into the graph. Its extraction instructions are shipped source, not external agent policy (SRC-1, src/compiler/ingest.ts):

const systemPrompt = You extract atomic, verifiable claims from source material. Each claim should be a single factual statement. Be precise and specific. Assign an initial confidence based on how well-supported the claim is by the source. Tag each claim with relevant topic tags. Identify related concepts (potential wiki page titles) for cross-linking. ---src/compiler/ingest.ts`

Local state survives individual commands: the CLI opens the same working-directory database rather than creating an isolated store per request (SRC-1, src/cli/context.ts, src/graph/store.ts):

export function openStore(): KnowledgeStore { return new KnowledgeStore(getDbPath()); } --- src/cli/context.ts

Later queries retrieve retained material before calling the model (SRC-1, src/graph/query.ts):

const { pages: relevantPages, claims: relevantClaims } = store.search( question, 50, ); --- src/graph/query.ts

Functional inclusions

• Production and admission: initialization, local-file and URL ingestion, directory ingestion, watched-file ingestion, claim extraction, comparison, resolution, entity-kind and metadata handling, confidence calculation and page summaries. These include embedded extraction, comparison and assignment prompts (SRC-1, src/cli/create.ts, src/cli/ingest.ts, src/cli/fetch-url.ts, src/cli/watch.ts, src/compiler/ingest.ts, src/compiler/diff.ts, src/compiler/resolve.ts, src/types.ts).
• Retention and maintenance: SQLite sources, pages, claims, provenance links, dependency and contradiction relations, events and full-text indexes; confidence updates, decay, cascade handling, lint/prune, health reporting, resurfacing, red-team review and discovery. Suggestions and critiques are included without assuming they are automatically applied (SRC-1, src/graph/store.ts, src/graph/temporal.ts, src/cli/metabolism.ts, src/cli/lint.ts, src/metabolism/health.ts, src/metabolism/redteam.ts, src/metabolism/resurface.ts, src/discovery/discover.ts).
• Reload and consumption: CLI queries and consensus queries, browser dashboard and its state-editing/bookmark routes, MCP read/write tools, exported library API and compiled/exported views. The optional query-answer save path writes a raw source file for later ingestion; it is part of the loop, not an excluded host responsibility (SRC-1, src/cli/query.ts, src/cli/save-answer.ts, src/compiler/consensus.ts, src/cli/serve.ts, src/mcp/server.ts, src/lib.ts, src/cli/compile.ts, src/render/markdown.ts).

The save path makes its later ingestion step explicit (SRC-1, src/cli/query.ts):

Run 'qw ingest ${filePath}' to compile it into knowledge, --- src/cli/query.ts

Tracked-tree coverage

The table accounts for every top-level tracked file or directory at SRC-1's commit. Included means inside the functional evidence boundary, not that every line has already received a detailed mechanism review. The initial inspection traced the CLI, library and MCP entry points through production, retention, maintenance and consumption. Bounded reading and symbol/call searches supplemented full reads. No material shipped path was inaccessible.

Frozen evidence and limits

The applicability cutoff is 2026-10-15. The reviewed boundary is exactly cfeecbffce5b6d0f16533b258a1e88edd86f731e, not a moving branch, npm release or live service. The acquired checkout's HEAD matched that commit and its working tree was clean when inspected. The supplied source identity remains https://github.com/anzal1/quicky-wiki; differing homepage/repository strings inside package.json are package-authored metadata, not authority to substitute another repository.

Overall evidence is code-grounded. SRC-1 supplies implementation and doctrine/design. No target code, tests, examples or model calls were executed. The tracked tree has no dedicated test/benchmark directory; package.json supplies build and typecheck scripts, while shipped lint, health, red-team and consensus paths provide internal checking/review mechanisms. This is not evidence of their accuracy or efficacy (SRC-1, package.json, .github/workflows/publish.yml, src/cli/lint.ts, src/metabolism/health.ts, src/metabolism/redteam.ts, src/compiler/consensus.ts).

External dependencies are Node.js, SQLite through better-sqlite3, SDKs and configured model endpoints, remote URL content, the browser's D3 resource, and downstream MCP clients or embedding applications (SRC-1, package.json, src/llm/adapter.ts, src/cli/fetch-url.ts, src/cli/serve.ts, src/lib.ts, src/mcp/server.ts). Their implementations and operation are deliberately outside this source boundary. This prevents conclusions about model quality, service reliability, dependency internals, fetched-document truth, browser delivery, client interoperability or enclosing-agent planning. Such host responsibilities are not attributed to Quicky Wiki.

User-created .quicky/ databases, raw corpora, generated wiki/ and dist/ outputs, deployed installations, execution traces and controlled experiments are not supplied as frozen evidence. Their absence prevents conclusions about actual retained contents, long-term knowledge improvement, production scale, confidence calibration, runtime safety or performance. Available source portions not yet inspected are follow-on inspection work, not access gaps and not exclusions from the registered Git revision.

Source register