The skills that keep the wiki current, the automation that runs it on a daily schedule, and the pipeline that turns canonical assets into live website pages.
Each skill has a narrow, well-defined job. Together they form the wiki lifecycle: ingest → synthesise → maintain → promote → retrieve.
The main ingest loop. A new file lands in the inbox — wiki-ingest runs the sensitivity gate, classifies it into one of 8 domains, synthesises edits to the concept pages, updates the domain index and log, archives the input, detects trend signals, and opens a draft PR for review.
Two ingest lanes: the repo lane (files dropped into inputs/inbox/) and the Drive lane (files uploaded to the Wiki Google Shared Drive Inbox — fetched via the Drive API, host-converted, archived to Drive on success).
Sensitivity gate (SCHEMA v0.10): class-scoped — public third-party scrapes bypass it; Wordnerds-generated or customer-touching material gets the full gate; ambiguous cases gate. --auto mode powers the daily run: clean files go end-to-end, anything needing a judgement call escalates to needs-review/.
Schema compliance audit. Checks frontmatter completeness, broken cross-references, stale-page warnings, and SCHEMA version alignment. Also runs as a PR-time hook — lint failures notify the Wiki Google Chat space and block auto-merge until resolved.
Scans concept pages for where wiki understanding has moved ahead of the canonical assets. Drafts proposed edits to positioning, methodology docs, or playbooks, then opens a PR. Deliberately human-run, not on the daily schedule — canon changes warrant considered review.
Reads trends-buffer.md across all domains. When a candidate trend hits the evidence threshold (SCHEMA §8), it drafts a synthesis and opens a PR to promote it into the canonical or concept layer, then posts a digest alert to the Wiki Chat space.
Retrieves relevant wiki knowledge on demand — called by other skills (the content pipeline, competitor-digest, outreach) as their "go fetch context" step. Searches concepts, canonicals, and index files; returns structured excerpts with sources: traceability. Reads only.
Two automated systems (added June 2026) that mean the wiki mostly manages itself. The human job has shifted from doing the ingest to reviewing the output.
Runs wiki-ingest --auto every morning at 09:30. It runs locally, so the laptop must be on — whatever it can see in the inbox gets processed.
Runs weekdays at 08:30 UTC. Applies lazy-consensus merge logic to skill-opened PRs: 96 hours open with no comment, it merges. Any comment holds the PR until a human resolves it.
A 7-stage pipeline in the wordnerds-site repo. wiki-query retrieves canon at stage 1; the Schema Composer produces a JSON schema that lib/renderer.ts renders to self-contained HTML; then stage 7, review-page, QAs the rendered page — strategy, copy/tone, AEO, SEO, conversion/UX balance and fact-check — and gates it (any BLOCKER → not ready). Only once it's clean does Pete deploy to Cloudflare via npm run deploy:<slug>.