Skip to content

index_llm wiki memory template

Jump to bottom
Priscila Saboia Moreira edited this page Jul 1, 2026 · 17 revisions

type: index up: "Home_llm-wiki-memory-template"

Index — llm-wiki-memory-template

Catalog of all wiki pages, organized by category.

Overview

  • Home: project summary and navigation
  • Schema: conventions and operations reference
  • Log: chronological record of wiki activity

Foundation

Mechanisms

Lessons learned

  • Lesson-Parallel-File-Drift: same content in two files
  • Lesson-Hook-Type-Matters: prompt-type hooks cannot be advisory
  • Lesson-Validation-Methodology: structural + behavioral testing
  • Lesson-Template-Contribution-Discipline: four universal rationalizations for AI agents contributing PRs to the template itself (single-writer default, scope creep, speculative PR numbers, state-locking test assertions). Sister page to Discipline-Gates at the contribution layer
  • Lesson-Gitignore-Anchor-Propagation: leading-slash .gitignore entries anchor to whatever repo they end up in; shipping a template-local block through ALWAYS_FILES silently shadowed derived projects' real CLAUDE.md and .claude/ paths (2026-06-18 bug report against template HEAD 80e2ee6, fix in PR #39)
  • Lesson-Bash32-Empty-Array-Hazard: on macOS bash 3.2 + set -u, "${arr[@]}" of a never-pushed empty declared array fires unbound variable; invisible on Linux bash 4+ so only surfaces when a fixture's grant pattern reaches the asymmetric empty-array case on macOS CI (2026-06-25, feature/adopt-sh Tier-1 fixture, fix in commit 57457db)

Open limitations

Applied examples

  • Applied-Example-Microelectronics-Tutor-Demo: wiki-llm as undergraduate tutor
  • Lessons-Learned-From-model_fusion: field report from a long Claude Code session in PaperAnalyticalDeviceND/model_fusion (NIR + PAD pharma analytics, 43 wiki pages, ~5 months of curation). Surfaces 12 implementation gaps verified against current template HEAD 88f4624 (2026-06-12): 0 fully addressed, 3 partial, 9 still missing. The separate "recursive bootstrap" finding is RESOLVED in PR #25 via scripts/instantiate.sh --dev-self
  • Analysis-of-model_fusion-Recommendations: assessment of the 12 recommendations above. Endorses #1, #4, #8, #9(b) as the high-value cluster; pushes back on #10 (variant proliferation), #6 (wrong mechanism — belongs in lint over the KG), #11 (overlaps with auto-compact), #2 (false-positive risk). First page filed under type: analysis, which is itself recommendation #4 — the page is empirical evidence for the gap
  • Analysis-of-Wiki-Push-Race-Pattern: reclassification of Multi-Agent-Write-Protocol from "aspirational" to "empirically needed, prototype shipped, wiring is the gap" based on a 2026-06-12 production collision in web_forager (3-4 human+agent teams; 20 commits of drift in the interim; log file conflict; ~3 min manual recovery). Recommends wiring wiki_push into the agent overlay's actual write path; rename the proto dir; pre-push hook as belt-and-suspenders. Second type: analysis page
  • OKF-Alignment-Ideas: working analysis of whether and how to align with Google's Open Knowledge Format v0.1. Includes empirical GitHub Wiki link-rendering results (only [[wikilink]] is existence-checked; OKF's [name](/path.md) form is broken in GitHub Wiki) and a recommendation across three vectors (frontmatter superset, drop reserved-filename namespacing once Obsidian aggregation moves to subdirectory-per-wiki, migrate body links to [[wikilinks]]). Third type: analysis page
  • Knowledge-Bundles-Framing: paper-brief synthesis written 2026-06-26. Reframes the template as an agentic knowledge-bundle system using OKF v0.1 terminology as the formalisation anchor; names the discipline layer (verification gate, write protocol, typed-edge ontology, KG, federation) as the contribution on top of the bundle format. Assumes the OKF-Alignment-Ideas recommendations are resolved; carries a current-implementation-status footer naming shipped vs forward
  • Cursor-Overlay-Validation: first live validation of the Cursor overlay on a real adopting project (2026-06-30); closes the README's "shipped but unvalidated" gap with empirical evidence — 4 behavioural tests pass (proactive read, link discipline without reminder, @wiki-experiment write end-to-end, honest reporting); 2 non-blocking observations nameable. Fourth type: analysis page

Governance and meta

Clone this wiki locally