Dineway Content Monitor

Dineway Content Monitor

Observe content without mutating it. Store bounded local evidence, record normalized observations through Content Pipeline MCP tools, and apply the active deterministic monitor policy. The plugin, not the Agent, owns tier transitions and deduplicated Fix triggers; the master Skill owns diagnosis, calendar priority, and the later Draft workflow.

Load references/ai-visibility.md and invoke dineway-content-ai-visibility before collecting or interpreting AI-engine observations.

Source boundaries

• Query GSC and other user-owned accounts through the user's local authenticated connector. Do not send those credentials to Forgeway.
• Observe ChatGPT, Perplexity, Claude, Gemini, Google AI Overviews, Grok, and DeepSeek through the user's local clients or Browser Use. These are client-owned compute, not Forgeway AI workloads.
• Crawl the first-party site locally; use Browser Use for rendered checks. Record internal links, orphan pages, duplicate topics, keyword conflict evidence, status, canonical, and structured data.
• Use Forgeway DataForSEO/Firecrawl only for managed API-key or infrastructure-scale evidence such as live SERP/rank/keyword datasets. Use dineway-seo-providers for bounded requests.

Every missing or failed source receives an explicit availability state. Never emit zero traffic, rank, citation, or issue counts in place of unavailable data.

Cadence

• Check new content during its first 30 days and other high-priority content every three days.
• Check stable content weekly and low-priority content monthly.
• Promote immediately after an anomaly. Demote only after three consecutive stable observations.
• Use the active versioned policy when it differs; record the policy version with evidence.

Workflow

1. Read content_pipeline_monitor_policy_get, due content_pipeline_monitor_state_list entries, native content identity, locale, revisions, publication time, linked observations, score snapshots, and prior Fix opportunities.
2. Collect only sources due for this tier. Preserve raw local evidence in .dineway/content/runs/<run-id>/jobs/<job-id>/monitor/evidence.json; normalize each result into a native observation with freshness, availability, artifact reference, dedupe key, and either numeric monitoring.metrics or explicit monitoring.integrity facts. Unavailable sources use an empty payload.
3. Compare like-for-like windows and sources. Distinguish measurement gaps, seasonality, query/SERP changes, site integrity failures, and genuine content decay as Agent analysis; do not convert that diagnosis into caller-authored thresholds.
4. Run the reviewed content_pipeline_monitor_observation_evaluate command once for each stored observation. Preserve the returned evaluation, policy, tier, next due time, and optional trigger in the local monitor receipt. Exact replay must reuse the stored evaluation; an older new observation fails closed.
5. Record an explainable new priority snapshot when current evidence materially changes demand, authority gap, freshness risk, effort, or publication readiness.
6. If evaluation returns a trigger, read it with content_pipeline_fix_trigger_get and use its linked Fix opportunity. Do not manually create another Fix. Repeated evidence reuses one active Fix opportunity; after that opportunity is terminal, later evidence may create new work.
7. Return to dineway-content-pipeline. Do not create a duplicate content item, mutate the page, or enter the Draft pipeline before the linked Fix opportunity is accepted.
8. Return the canonical Monitor artifact and typed payload linked to observations, deterministic evaluation receipts, policy version, next due time, and optional Fix trigger.
9. Let the master call content_pipeline_stage_complete, which derives the receipt, completes the Attempt/Assignment, and creates/accepts the immutable Monitor Result. Do not call granular Result operations.

Output

Return source availability, observation IDs, comparable changes, anomaly classification, monitoring tier decision, policy/evaluation/trigger IDs, score-snapshot rationale, linked Fix decision, and next due check. If evidence is inconclusive, say so and retain the policy-derived next due time rather than manufacturing certainty.