identity kind path revision sha256

LLM Project Wiki boundary

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

Boundary and evidence

The target is the complete shipped prompt package for creating, maintaining and consulting a project-specific Obsidian wiki through Claude Code. Its intended use is to give later model sessions a persistent project reference instead of repeatedly reading raw code. It is in scope as a memory/knowledge/context-engineering system: its bootstrap, interpretation of sources, wiki writing and later lookup depend on the host model following instructions (SRC-1, README.md, SETUP_PROMPT.md).

A Claude Code prompt that bootstraps a self-maintaining Obsidian knowledge base for any software project, and wires it as the primary reference for all future LLM work on that project. --- README.md

The boundary is whole-system for this shipped prompt package, not for a generated project installation or the enclosing Claude Code runtime. The tracked tree contains only README.md and SETUP_PROMPT.md; both were inspected completely at commit 1fdfba9129a46afd18aa68a83f12d6716ab694c6. The applicability cutoff is 2026-10-14. Evidence is doc-grounded: the repository supplies instructions and embedded code examples, not a materialized vault, executable maintenance package or inspectable run trace (SRC-1, README.md, SETUP_PROMPT.md).

Shipped-path coverage

Top-level tracked path Coverage Role and material scope Limit or conclusion prevented
README.md Included; completely inspected Usage entry point; new-project and existing-vault invocation; ongoing sync and ingest instructions; declared wiki-first consumption flow; attributed example output and supported-stack claims. Example page counts and claimed capabilities are reported operation or declared intent, not demonstrated outcomes.
SETUP_PROMPT.md Included; completely inspected Step 0 branches between bootstrap and existing-vault audit. Phases 0–3 specify project inspection, persistent vault layers, schema, templates, index, log, health dashboard, source sync and diff-based ingest. Phase 4 specifies later callers through project instruction files, status commands and phase-completion commands. Phases 5–6 specify tracking, manuals, logging and self-audit. The gap heuristic, quality bar and Phase G specify coverage, link/schema checks, source refresh, protocol repair and audit logging. These are instructions for model-mediated generation and maintenance. They do not establish that generated scripts, commands or future host sessions actually satisfy them.

Functional inclusions cover the entire declared knowledge loop: producing wiki pages from project sources; retaining sources, compiled pages and activity records; selecting pages through the index; marking pages stale after code edits; updating pages through sync and diff-based ingest; resolving logged gaps; and auditing an existing vault. The specified generated callers and support files are part of the design being analyzed, even though their concrete project-specific realizations are not shipped here (SRC-1, SETUP_PROMPT.md).

The governing document. A future LLM maintaining this wiki must read it first. It must define: --- SETUP_PROMPT.md

  1. Read changed source → read wiki page → update only changed sections → bump updated:
  2. Check log.md for open [gap] entries and resolve them
  3. Create new wiki pages for any new entities introduced by the changed sources

  4. Append ingest entry to log.md --- SETUP_PROMPT.md

  5. Open <vault path>/index.md — use the Semantic Lookup table to navigate

  6. Read the relevant Wiki/ page(s) — they contain full code excerpts and cross-references
  7. Only if the wiki is insufficient: fall back to existing memory files
  8. Only for a targeted edit: open the actual source file --- SETUP_PROMPT.md

External dependencies are Claude Code's model, filesystem and command execution capabilities; optional Gemini instruction-file integration; the consuming project's source, memory, rules and Git history; shell utilities including rsync, grep and Git; and Obsidian with Dataview and Templater for the specified vault interface. The prompt describes these dependencies but does not ship their implementations (SRC-1, README.md, SETUP_PROMPT.md).

The enclosing host runtime is deliberately excluded: this evidence cannot establish its instruction loading, slash-command execution, context limits, scheduling or enforcement of wiki-first behavior. Consuming project code and generated vault artifacts are unavailable in this source set: the analysis cannot establish actual entity coverage, freshness, retrieval accuracy, generated-script correctness or successful persistence and reload across sessions. Dataview, Templater, Obsidian and Gemini implementations are excluded: their rendering, plugin compatibility and instruction-following behavior cannot be inferred from the prompt. The linked Karpathy gist is not part of the frozen source and was not inspected; attribution in the README does not support independent conclusions about that external proposal. No target execution or experiment was performed; effectiveness and causal improvement remain unestablished. These limits do not remove any shipped path from the prompt-package boundary (SRC-1, README.md, SETUP_PROMPT.md).

Source register

Source ID Kind Identity/location Revision or capture Evidence layer Inspected scope Citation anchors Access gaps and conclusion prevented
SRC-1 git https://github.com/akash-r34/llm-project-wiki 1fdfba9129a46afd18aa68a83f12d6716ab694c6 doctrine/design; reported operation Access root: /home/zby/llm/commonplace/.commonplace/worktrees/llm-project-wiki-c6c664168f3e/related-systems/akash-r34--llm-project-wiki. Verified HEAD and enumerated all tracked paths. Doctrine/design: complete README.md and SETUP_PROMPT.md, including embedded sync/audit examples and all bootstrap, maintenance, consumption and repair instructions. Reported operation: README's example project and page-count account. Doctrine/design: README, setup prompt. Reported operation: the README's “What the Wiki Contains” section at the same pinned anchor. All tracked files inspected; no uninspected material shipped path. No concrete generated installation, host implementation, execution trace, evaluator results or causal experiment is supplied. Instructions support declared operation, not verified enforcement or effectiveness; reported counts do not establish a reproducible run.