lint: retire ALWAYS_FILES-era claims from contract pages (manifest consolidation PR #60, doc fixes PR #67)
Add Cursor-Overlay-Validation analysis page (first live validation, 2026-06-30)
Filed as fourth type: analysis page. Closes the README Status gap that
read 'Cursor overlay is shipped but unvalidated in a live session' with
empirical evidence from a real adopting host: rules load automatically
(wiki-as-memory.mdc + CLAUDE.md always-applied), link discipline applied
without explicit reminder, @wiki-experiment write produces page + index
+ log with two-commit rule honoured, honest reporting validated against
gh pr view.
Two non-blocking observations recorded:
1. Commit-without-asking is the designed behaviour per lw:wiki-maintenance
(stop-and-confirm gates live at PUSH, not COMMIT).
2. Rhetorical 'prior snapshot' framing without an actual prior on disk
is worth tightening in .cursor/rules/wiki-experiment.mdc spec.
Bearing on related work: empirical evidence narrows issue #65 to 'adopt
entry-point only — overlay itself validated'; comment posted on #65
with concrete 3-part fix sketch.
Index + Home updated under Applied examples category.
Knowledge-Bundles-Framing — paper-brief synthesis (OKF + discipline)
Filed for handoff to the agent writing the llm-wiki-paper.
Reframes the template as an agentic knowledge-bundle system,
using OKF v0.1 terminology as the formalisation anchor and
naming the discipline layer (typed edges, verification gate,
write protocol, KG, federation) as the contribution on top of
the bundle format.
Structure: TL;DR; OKF-anchored terminology; agentic stance;
five-step bundle lifecycle (create/adopt/maintain/federate/
compose); where we extend OKF; cross-cutting properties; honest
limits; naming and positioning for the paper; current
implementation status footer separating shipped vs forward as
of 2026-06-26.
Assumes OKF-Alignment-Ideas recommendations are resolved
(frontmatter superset, OKF-canonical reserved filenames,
[[wikilinks]] body links, posture (c) companion spec for typed
edges). The implementation-status footer makes the forward-
state assumption auditable.
Pages touched:
- Knowledge-Bundles-Framing.md (new, type: synthesis)
- index_llm-wiki-memory-template.md (Applied examples entry)
- Home_llm-wiki-memory-template.md (Applied examples entry)
- 9 See-also back-refs across substrate pages
wiki: record advisory-abort redesign and bash 3.2 hazard from feature/adopt-sh
Two ingest entries from the Tier-1 fixture pass on 2026-06-25
(branch feature/adopt-sh HEAD 57457db):
- Adopt-Existing-Repo-Design.md: new 'Decision: advisory abort when
host is already adopted' section. Composite detector (Signals A/B/C,
2-of-3 threshold), --force escape hatch, empirical classifications
against 6 real repos.
- Lesson-Bash32-Empty-Array-Hazard.md: new page. macOS bash 3.2 +
set -u + empty array hazard, the asymmetric TOUCH_MISSING fixture
that surfaced it, bash -x trace pinpointing scripts/adopt.sh:434,
single-site fix in commit 57457db, audit of 5 sibling scripts.
Index updated with one-line hooks for both pages. Log entry filed.
Co-authored-by: Priscila Saboia Moreira <pmoreira@nd.edu>
OKF-Alignment-Ideas — working analysis vs Google's Open Knowledge Format v0.1
Reads the actual SPEC.md (not just the blog post) and folds in
empirical GitHub Wiki link-rendering test results from a
temporary push-then-revert probe. Recommends three changes
(frontmatter superset; drop reserved-filename namespacing once
Obsidian aggregation moves to subdirectory-per-wiki; migrate body
links to [[wikilinks]]) and defers/rejects five others, with the
gate decision pinned to the user's call on the Obsidian-vault
layout.
Pages touched:
- OKF-Alignment-Ideas.md (new, type: analysis, third in the wiki)
- index_llm-wiki-memory-template.md (Applied examples entry)
- Home_llm-wiki-memory-template.md (Applied examples entry)
- Typed-Edges-and-Frontmatter.md (See also back-ref)
- Wiki-LLM-Pattern.md (See also back-ref)
- Sync-Flow.md (See also back-ref)
- Knowledge-Graph-Pipeline.md (See also back-ref)
- Lesson-Parallel-File-Drift.md (See also back-ref)
ingest: Adopt-Existing-Repo-Design
New synthesis page consolidating issue #6 discussion (Priscila's original
body + @hegu-1's two follow-up comments) into a concrete design for
scripts/adopt.sh.
Key reframe from hegu-1 captured: never-touch is computed, not enumerated.
Three lists model — ADD allowlist (template-owned), TOUCH grants
(host-owned via .llm-wiki-adopt-grants.yml), NEVER-TOUCH (derived).
Test approach proposes a controlled fixture + a real-repo dry-run against
crcresearch/FUNSD as a no-wiki public candidate.
Index updated under Mechanisms (adjacent to other in-design pages).
Lesson-Gitignore-Anchor-Propagation — sync-time footgun from PR #39
`.gitignore` leading-slash entries anchor to the repo root of whichever
clone the file lives in. Shipping a template-local block through
`ALWAYS_FILES` moved the anchor into derived repos, where it silently
shadowed their real paths. Mostly latent (tracked files unaffected) but
bites: projects that legitimately track `.claude/hooks/`, paths that are
untracked at sync time, and contributors confused by `git check-ignore`
reporting their own `CLAUDE.md` as ignored.
Root cause classified as "scope-sensitive content shipped through a
scope-preserving channel" — sister to Lesson-Parallel-File-Drift (same
content in two files) but distinct (same file in two scopes).
Pages touched:
- Lesson-Gitignore-Anchor-Propagation.md (new)
- Home_llm-wiki-memory-template.md (Lessons learned entry)
- index_llm-wiki-memory-template.md (Lessons learned entry)
- Sync-Flow.md (See also back-ref)
- Lesson-Parallel-File-Drift.md (See also back-ref)
- Lesson-Template-Contribution-Discipline.md (See also back-ref)
ingest: Lesson-Template-Contribution-Discipline
New synthesis page filed in response to user observation that lessons
learned during template development should live in wiki memory rather
than per-user Claude local memory, so they survive a clean pull (via
--dev-self per PR #25).
Captures four rationalizations and their empirical incidents:
1. Defaulting to single-writer when web_forager proved multi-writer
(incident: 2026-06-12 Omegaice/csweet1 log_*.md collision, ~3 min
manual reconcile).
2. Introducing speculative "Option C" integrations the user did not
propose (MCP overreach in PR7's first draft, pushed back on
2026-06-13).
3. Hardcoding speculative PR numbers in code (PR #28 stale #26
references; Priscila cleaned up four in commit e48f7d4).
4. Adding structural assertions that lock historical state rather
than catch behaviour regressions (PR6's three rename-locking
assertions, flagged 2026-06-13).
Each pattern has the same shape as Discipline-Gates' Universal
Rationalizations, just one layer up: instead of rationalizing a bad
wiki entry, the agent rationalizes a bad PR draft, a noisy test, or
a stale reference. Discipline-Gates See-also gains a back-reference.
Sister page to Lesson-Hook-Type-Matters, Lesson-Parallel-File-Drift,
Lesson-Validation-Methodology — matches their synthesis-type +
empirical-motivation style.
After this lands the four corresponding personal Claude memory entries
can be deleted; the lessons now live where they survive a clean pull.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
ingest: Analysis-of-Wiki-Push-Race-Pattern + reclassify Multi-Agent-Write-Protocol
New type: analysis page filed in response to a 2026-06-12 production
collision in wiki/web_forager.wiki/ (3-4 human+agent teams; Csweet1
push rejected after Omegaice advanced remote by 20 commits in the
interim; conflict on log_web_forager.md from same-date entries from
both authors; ~3 minute manual rebase recovery).
The Multi-Agent-Write-Protocol prototype at
scripts/multi-agent-write-protocol-proto/ (shipped in PR #11; 9/9 CI
scenarios green on ubuntu-latest + macos-latest) handles exactly this
collision shape automatically via wiki_push's mechanical union-merge
for log_* files. The prototype is implemented but not wired into any
agent overlay; the only callers are the 9 CI scenarios.
Reclassification: the Multi-Agent-Write-Protocol banner is rewritten
from "design + prototype only, not yet wired" to "design + prototype
shipped, production wiring is the gap." Implementation-Status's row is
updated to name the wiring (not the design) as the specific gap and
to outline what a wiring PR should contain.
Second type: analysis page filed (after Analysis-of-model_fusion-
Recommendations earlier today). Both acknowledge the SCHEMA violation
that PR #29 closes.
Cross-references: Multi-Agent-Write-Protocol gains See-also entry to
the new page; Analysis-of-model_fusion-Recommendations and
Lesson-Validation-Methodology gain back-refs as sister/related pages.
Index updated under Applied examples (after the sister analysis page).
Home updated per PR #32's category-level rule: the page is significant
enough to warrant a Home entry, not just an Index entry.
Wiki-Corrections-Log entry appended for the receipt.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
ingest: Analysis-of-model_fusion-Recommendations (first type: analysis page)
Assessment of the 12 recommendations in Lessons-Learned-From-model_fusion,
filed in response to user's question about which would bring the greatest
value. Endorses #1 SessionStart auto-load, #4 analysis/decision page
types, #8 Claude-memory boundary, #9(b) reciprocal-edge enforcement.
Pushes back on #10 (variant proliferation risk), #6 (belongs in lint
over the KG, not a hook), #11 (overlaps with auto-compact), #2 (false-
positive risk).
Dogfood: filed under type: analysis, which is not declared in SCHEMA --
the point being that this is the exact gap recommendation #4 names. The
page is empirical evidence for the recommendation it endorses. SCHEMA
violation acknowledged in the page body.
Index updated under Applied examples. Lessons-Learned-From-model_fusion
See-also gains the reciprocal back-reference.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
lint: KG aspirational -> implemented; bidirectional back-references; example cleanup
Stale-claim correction: PR #14 shipped scripts/kg/ (rdflib + pyshacl
in-process KG pipeline) to template main, but nine wiki pages still
described the pipeline as aspirational. Reclassified Knowledge-Graph-
Pipeline to implemented; corrected architecture from Fuseki to rdflib +
pyshacl; flagged the distribution gap (scripts/kg/ not yet in
ALWAYS_FILES, likely to ship as features/kg/ under RFC #13). Updated
Implementation-Status, Wiki-LLM-Pattern, Three-Operations, Typed-Edges-
and-Frontmatter, SCHEMA, Limitation-Inverse-Edges-Missing, Sync-Flow,
and the index to match.
Example cleanup: Wiki-LLM-Pattern line 23 used the placeholder example
[Markov-Chain-v5c2](Markov-Chain-v5c2). Replaced with a real wiki page
name so lint and the KG see a valid edge.
Bidirectional back-references added for the high-value concept <-> lesson
/ limitation pairs (Verification-Gate, Discipline-Gates, Lesson-
Validation-Methodology, Lesson-Parallel-File-Drift, Template-vs-Derived-
Projects, PostToolUse-Hook, Agent-Overlays). Index/log roll-ups
(Implementation-Status, Wiki-Corrections-Log) intentionally left one-way.
Wiki-Corrections-Log entry appended with the receipt.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
ingest: Lessons-Learned-From-model_fusion (field report + 12-gap verification table)
New synthesis page filed as a sister to Applied-Example-Microelectronics-Tutor-Demo:
a second empirical test of the template from a different domain (NIR + PAD pharma
analytics, 43 wiki pages, ~5 months of active curation in
PaperAnalyticalDeviceND/model_fusion).
Headline content:
- Verification table of 12 implementation-gap recommendations against current
template HEAD fc2472a: 0 fully addressed, 2 partial, 10 still missing.
- Each recommendation labelled with empirical motivation (what fell through the
cracks in the source session) and a refined fix.
- Cross-references existing template lessons (hook-type-matters,
parallel-file-drift) and the inverse-edges-missing limitation that
recommendation #9 builds on.
- Includes the "recursive bootstrap problem": cloning the template lands
contributors without llm-wiki context. Proposed fix: dev-mode bootstrap
that clones the template's own GitHub Wiki locally, gitignored and
excluded from the instantiate allowlist.
Updates the index Applied-examples section and appends a log entry.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Ingest Multi-Agent-Write-Protocol spec page
A new aspirational page documenting an OCC-with-semantic-retry write
protocol for wikis with multiple LLM-agent writers. The protocol prevents
push-level collisions by construction: each agent writes via its own
contribution branch, attempts a merge against current main, and either
pushes cleanly or re-plans its writes semantically before retrying. No
separate reconciler is needed; the writing agent reconciles its own
attempts.
The page covers: the protocol's six steps; conflict classification
(union-merge for index/log files via .gitattributes, three-way merge for
non-overlapping content, semantic resolution for overlapping content);
required agent capabilities; failure modes (livelock cap K=3, network
faults, stale branches, name collisions); four worked examples spanning
the mechanical-merge and semantic-resolution cases; open questions; and
cross-references to existing wiki pages (Verification-Gate, Sync-Flow,
PostToolUse-Hook, etc).
Implementation-Status gains the page under Aspirational, pointing at the
deterministic prototype in progress at the Omniscient_2 sister repo.
Home and index updated in the Mechanisms section.
Log entry follows in a separate commit per Log-Entry-Attribution.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Ingest Test-Harness page; add In review bucket to Implementation-Status
Documents scripts/test-mvp/ and the GitHub Actions workflow that ship in
PR #10 (chrissweet/llm-wiki-memory-template add-test-harness), addressing
template issue #5. The harness implements the structural half of
Lesson-Validation-Methodology: smoke against real template, e2e against a
simulated derivative with MVP patches applied. 91 assertions across 1
smoke + 6 e2e stages on bash 3.2-compatible code, matrix on ubuntu and
macOS.
The page makes the gap-analysis explicit: the harness covers stage 2,
stage 5, and Agent-Instance assertions in stage 1 that the trimmed MVP
slates for deferral; the harness does NOT yet cover ALWAYS_FILES updates
to the sync scripts (the footgun Sync-Flow documents and PR #1 hit).
Filed as follow-up work, not yet implemented in the harness.
Implementation-Status gains a fourth bucket ("In review") for features
that exist as code in an open PR but are not yet merged into template
main. Home and index updated in the Mechanisms section.
Log entry to follow in a separate commit per Log-Entry-Attribution
convention.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Flag KG pipeline aspirational; correct none-overlay framing; fix Sync-Flow ALWAYS_FILES; add Implementation-Status + Wiki-Corrections-Log
Reconciles the wiki with the template code on main, where notes-driven
documentation had drifted from what shipped:
- Knowledge-Graph-Pipeline is described across 8 pages as if implemented,
but scripts/kg/build-graph.sh does not exist in the template repo. The
source page now carries a NOT YET IMPLEMENTED banner and a
status: aspirational frontmatter field. Citations on the seven dependent
pages are softened to "designed, not yet implemented" so the aspirational
status is visible at every mention.
- The none "overlay" is not an overlay: --agent=none in instantiate.sh is
a no-install flag, and there is no wiki/agents/none/ directory.
Limitation-Untested-Overlays and Agent-Overlays are reframed accordingly,
separating minimal mode from the two true overlays (claude-code, cursor).
Limitation-Untested-Overlays also corrects a phrasing claim: the README
uses "Shipped but not yet validated" / "Likewise unvalidated", not
"untested-live".
- Sync-Flow's ALWAYS_FILES section was inaccurate: it mixed contents from
four separate arrays and listed the nonexistent scripts/kg/*.sh. Section
rewritten to describe all five arrays (ALWAYS_FILES, CLAUDE_FILES,
CURSOR_FILES, ONE_SHOT_FILES, SUBSTITUTE_FILES) with the file lists
actually present in the sync scripts on main.
New pages:
- Implementation-Status.md: single normative source for what is real today
vs aspirational, with a Partially Implemented bucket for Cursor's
structural-but-not-behavioral validation.
- Wiki-Corrections-Log.md: append-only log of each correction; this commit
is the first entry.
Home and index updated with a "Governance and meta" section linking the
two new pages.
Verification debts noted in the Corrections Log but not corrected this
round: Applied-Example-Microelectronics-Tutor-Demo staleness, the
permission-model claims in Limitation-Prompt-Injection-Residual, and the
existence of chrissweet/llm-wiki-tutor-template cited by Governance.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
ingest: Governance — three-tier model for variants
A new Foundation page documenting where features belong as the wiki-llm
pattern spreads from base template into specialized variants (e.g.
chrissweet/llm-wiki-tutor-template, future codebase-onboarding and
grant-authoring siblings).
Extends Template-vs-Derived-Projects by introducing a third tier
between the base and an instantiated project: the variant template.
The two-layer model in Template-vs-Derived-Projects compresses variant
+ base into "the template" — once a variant exists, the distinction
matters because feature placement decisions cascade differently across
three tiers than two.
The page covers:
- The placement rule for new features ("would a project unrelated to
this variant's behavior still benefit?") with a table mapping base
vs variant vs derived project, and examples from the current
ecosystem.
- Variant-fork vs feature-flag criteria, including the canonical
example of a grant-authoring variant whose "write it for me" rule
would conflict with a tutor variant's "never deliver the answer."
- Rebase discipline referencing Sync-Flow's ALWAYS_FILES contract,
escalated one tier (variants rebase from base the same way derived
projects rebase from variants).
- When a feature graduates from variant to base (two variants
implementing the same thing, or a base-level user asking).
- Ownership / drift detection so variants don't silently reimplement
each other's wheels.
- Three open questions for resolution as the variant population grows:
variant discovery, versioning, cross-variant features.
Cross-references Template-vs-Derived-Projects, Sync-Flow, Applied-
Example-Microelectronics-Tutor-Demo, and Lesson-Parallel-File-Drift
(the same footgun, escalated to the variant tier).
Home and index updated under Foundation, adjacent to Template-vs-
Derived-Projects.
Initialize wiki: scaffold + first ingest of template knowledge
Scaffolded SCHEMA, Home, index, log, and Home.md redirect to match the
canonical create-mode output of wiki/init-wiki.sh. Authored 18 content
pages across five categories (foundation, mechanisms, lessons,
limitations, applied examples) capturing the wiki-llm pattern, the
template's structure, the contents of PR #1 (discipline gates,
verification gate, PostToolUse hook) and PR #4 (log entry
attribution), the parallel-file-drift / hook-type / validation-method
lessons surfaced by PR review, and the three open limitations
(prompt-injection residual, missing edge inverses per issue #3,
untested non-Claude-Code overlays).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>