auto-analysis
Placeholder. One reader per system a session touches (git, deployments, cloud logs, trackers, billing), each simple and independent. They pull what happened there into records next to ../02-measure/'s, so ../03-analyze/ can join them with sessions. Each file lists coverage (first, next, long tail), workflow and pitfalls.
- Input: the external system's API or local files.
- Output: records keyed by user, time and repo or project.
- Data it touches: metadata only by default (who, when, what changed, status), never content unless the owner enables it. Each file states exactly what it reads. Credentials are read-only.
| File | Systems (first) | Used for |
|---|---|---|
git.md |
GitHub, GitLab | spend per merged change |
deployments.md |
GitHub Actions, GitLab CI | do agent changes pass; headless agent runs |
issue-trackers.md |
Jira, GitHub Issues, Linear | intent and business impact |
billing.md |
Anthropic, OpenAI, cloud billing, seats | real spend vs telemetry estimate |
cloud-logs.md |
Datadog, Grafana, OTel backends | infra and service impact |
mcp-usage.md |
harness MCP configs and call telemetry | MCP effectiveness |
docs.md |
Confluence, Notion, Google Docs, repos | non-technical users, org visibility |
chat-usage.md |
ChatGPT, Claude, Gemini, M365 Copilot | non-technical users |
quality.md |
linters, scanners, benchmarks | security, performance, usability of agent changes |
Corporate login and user to team mapping: ../06-optional/identity.md.