hoo-weekly

Weekly check

Resolve the brain: use registry.get_active() when running interactively; scheduled runs receive the brain path from the routine configuration.

1. Run core.contracts.check_schema(brain_path) first. If not compatible, relay the action string and stop before any of the steps below.

2. Read profile. Launch analytics-reporting-chief and serp-ai-monitor agents (parallel) with the profile path.

3. Save their reports under runs/YYYYMMDD-weekly/ (01-analytics.md, 02-serp-ai.md, REPORT.md synthesis).

4. Append the week's headline observations as signals. Include the week-over-week ai_referrals trend read from the week's daily signal lines (the AI-surface list is maintained in hoo-daily's SKILL.md - quote it, never extend it here); days without an ai_referrals line are stated as gaps, never interpolated. A sustained rise on a single landing page (3 or more of the week's daily lines) feeds the citation tracker: name the page in REPORT.md so the next citation run checks whether a new AI citation explains the traffic. 3.5. Stage classification. Build rows from this run's GSC page/query pull (one dict per query: clicks, impressions, position, plus query and page where the pull has them) and call core.stage.classify(rows): PYTHONPATH="$CLAUDE_PLUGIN_ROOT/lib" python3 -c "..." importing core.stage. The thresholds live in that module and are never restated here (see docs/INFORMATION-MAP.md in the repo). Record the result as ONE structured signal line - the exact token stage: <early|growing|established> followed by the returned reason sentence - and repeat it as the first line of REPORT.md. No GSC pull this run means no stage: line at all, never a guessed one. GROWING or ESTABLISHED: every section below runs exactly as it runs today; this step changes nothing for a site that has clicks. EARLY: run the early-stage report below in place of the three volume-gated detectors.

5. Invoke the organic-os:hoo-reflector skill to propose skillbook deltas from this week's signals + outcomes.

6. Queue any proposed work items; rebuild queue; notify per approval channel with one call: PYTHONPATH="$CLAUDE_PLUGIN_ROOT/lib" python3 -c "..." importing core.approval and calling notify_pending(<brain>, send), where send delivers over the configured channel (same channel selection as skills/onsite-propose step 4). An item is marked notified ONLY after its send returns without raising, so a failed send is retried on the next run rather than lost. Do not call is_notified or mark_notified by hand. Commit + push if git.

Any content brief this run emits for comparison-intent queries (vs, alternative, best X for) sets brief_type="comparison" on create_item; absence means explainer (see docs/site-repo-contract.md).

Run cost: wrapper-invoked runs land one row in ~/.config/organic-os/cost-ledger-YYYYMM.tsv (written by the runtime wrapper, never by this skill); the Monday report is what surfaces it.