# Central Agentic Ops Dashboard

> Agent entry point for the published CAO operational dashboard. This file routes agents to existing skills and data surfaces; it is not a data source or specification.

- Human dashboard: [open /cao/](./)
- Broader CAO documentation: [Central Agentic Ops documentation](../)
- Agent analysis guide: [choose CLI or MCP](../agent-analysis/)

## Choose the cheapest access path

1. Read this file.
2. For a procedure, read the matching raw skill.
3. For a cheap common metric, read `agent-summary.json`.
4. For one campaign, workflow, run, repository, or session, use the named-query MCP tools.
5. For large joins, history, aggregates, or exploratory SQL, download the SQLite snapshot with the CAO CLI.
6. Use the dashboard UI only for interactive human visualization; do not scrape it.

## Skills

- [Set up CAO](https://raw.githubusercontent.com/githubnext/gh-cao/main/skills/setup-cao/SKILL.md): Bootstrap a control plane, repository scope, and authentication.
- [Diagnose CAO](https://raw.githubusercontent.com/githubnext/gh-cao/main/skills/debug-cao/SKILL.md): Investigate policy, credential, workflow, Activity, dashboard, or safe-output failures.
- [Analyze CAO](https://raw.githubusercontent.com/githubnext/gh-cao/main/skills/analyze-cao/SKILL.md): Choose an access path and analyze deployed Activity evidence.
- [Use the CAO CLI](https://raw.githubusercontent.com/githubnext/gh-cao/main/skills/cao-cli/SKILL.md): Download, inspect, and query the SQLite snapshot.
- [Create a campaign](https://raw.githubusercontent.com/githubnext/gh-cao/main/skills/create-cao-campaign/SKILL.md): Author a new governed orchestrator and worker campaign.
- [Add a campaign](https://raw.githubusercontent.com/githubnext/gh-cao/main/skills/add-cao-campaign/SKILL.md): Discover, compare, and install an existing catalog campaign.

## Published Activity artifacts

- [Agent summary](./agent-summary.json): generated from the Activity inventory; schema version 1; 942 bytes; public; freshness is in `generatedAt` and per-source metadata. Answers campaign totals, disabled campaign names/counts, workflow state counts, repository count, and source freshness.
- [Inventory sources](./inventory-sources.json): generated by Activity repository and workflow discovery; Dashboard Language source envelopes; 1563325 bytes; public; freshness is in each source's `metadata.as-of`, `completeness`, and `freshness`. Answers detailed campaign, repository, workflow, marketplace, and policy inventory questions; prefer the summary or MCP when possible.
- [Payload hash manifest](./payload-hashes.json): generated during Activity/dashboard publication; name-to-SHA-256 mapping; 56465 bytes; public; identifies the exact SQLite and compacted shard payloads but is not freshness metadata. Answers which payloads belong to this snapshot and their integrity digests.
- Compacted run shards under `./gh-aw-logs-runs/`: generated by Activity compaction; JSONL run identity and aggregate records; 17 files, 6810927 bytes; public through names in the payload manifest; timestamps on records provide observation time. Prefer MCP or SQLite rather than downloading every shard.
- Compacted record shards under `./gh-aw-logs-records/`: generated by Activity compaction; JSONL domain, tool, audit, issue, evaluation, and related records; 413 files, 426291768 bytes; public through names in the payload manifest; timestamps on records provide observation time. Use only for bounded ingestion, not ad hoc browsing.
- [Activity SQLite](./gh-aw-logs.sqlite): rebuildable canonical query projection; 758366208 bytes; public. Use the CAO CLI download flow for integrity verification and bulk analysis.

## Targeted queries with MCP

- Prefer MCP for one campaign, one run, recent failures for one workflow, or a specific repository, workflow, or session.
- Call `cao_catalog` with `kind: pages` or `kind: queries` to discover reviewed dashboard queries.
- Call `cao_query` with a discovered query `id`, declared `parameters`, and a bounded `limit`.
- MCP exposes no raw SQL or writes. Preserve result `availability`, `completeness`, `freshness`, and `as-of` metadata.
- Setup and transport details: [Agent analysis](../agent-analysis/).

## Bulk analysis with CAO CLI and SQLite

1. Install or use the repository-local CAO CLI.
2. Run `cao download --url https://githubnext.github.io/gh-cao/cao/payload-hashes.json`.
3. Query `.cao/gh-aw-logs.sqlite` with `cao query`, reviewed named queries, or local SQLite tooling.
4. Use this route for large joins, historical analysis, aggregate computation, exploratory SQL, or analysis across many repositories and runs.

Do not issue thousands of MCP calls when one local SQLite analysis is cheaper.

## Schema and freshness

- [Authoritative dashboard data model](https://github.com/githubnext/gh-cao/blob/main/specs/dashboard-data.md) defines canonical entities, storage, and freshness semantics.
- [Dashboard data model guide](../dashboard-data-model/) summarizes repositories, workflows, runs, jobs, sessions, events, transactions, safe outputs, graders/evals, and campaign projections.
- Read `agent-summary.json.generatedAt` for summary generation time and its `freshness` entries for source state.
- Named CLI/MCP query results carry `availability`, `completeness`, `freshness`, and `as-of`; do not treat missing, stale, partial, or unavailable evidence as zero.
- The payload manifest binds the published snapshot by SHA-256. File age alone does not prove source freshness.