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)

chrissweet committed Jun 18, 2026

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>

chrissweet committed Jun 13, 2026