Standard operating procedures
Agent-facing procedures. Do not duplicate them in /docs; link here when a guide needs the steps.
| SOP | Use when |
|---|---|
| Eval-driven development (alpha) | Writing or gating agent evals |
| EDD production telemetry | Turning live misses into cases |
| Behavior catalog and XFN | Inventorying tests before a change |
| Context budget | Always-on files are too large |
| MCP library | Composing one MCP profile |
| Hypothesis-driven development | Product bets, experiments, feature flags, kill criteria |
| Hypothesis-driven debug | A failure needs RCA, not more logging |
| API contracts | Published HTTP or event contracts |
| DB migration | Expand/contract schema changes |
| Release | Shipping with a conventional title |
| Conventional commits | Commit message output (ticket id when in play); stay on main, uncommitted |
| Linear ticket execution | Claim In Progress, assign host agent, commit message with id |
| Host subagent launch | Parent uses wk agents launch-prompt; handover is the contract |
| TDD Guard hooks | Cursor/Claude red-before-green (wk tdd-guard) |
| External skills | Pinning upstream skills |
| Complexity hotspots | Pruning a hot path, or a git crime-scene report → Linear after a human gate |
| Cloudflare analytics ops | RUM / beacon diagnosis |
| PostHog product analytics | Cookieless SDK, privacy notice, official MCP |
| Product signal intake | Two-session PostHog findings → human gate → Linear |
| SonarQube findings | Triage Sonar issues; skip kit disagreements (Actions stay @vN); then fix or file one typed ticket |
| Coding-agent observability | Token usage, cost, and OTel traces in community SigNoz Docker or SigNoz Cloud (Cloud Agents); query via wk mcp signoz or hosted MCP; wk eval stays the quality gate |
| Quality loops through Linear | CI / RUM / scouts → typed tickets → hygiene → work picker draft PR; Dependabot / CodeQL stay on vendor PRs |
Operator narrative: What Waykit gives you. Nested loops and outside sources: Quality loops.