Appearance
Workflows
Higher-level pipelines the harness runs. Each node links to its skill; dotted arrows are feedback loops — where a step sends you back to revise and re-run.
Idea → shipped
A fuzzy idea — shaped, red-teamed, sliced into placed issues, then built.
brainstorm → check-reasoning → to-issues → triage → organise → build-sprint
- ↩
check-reasoning→brainstorm— FAIL: revise spec, re-check
QA PR & Learn
Everything green before a PR — then capture what the work taught.
qa-tests → qa-code → check-reasoning → find-untracked-work → find-learnings
- ↩
check-reasoning→qa-tests— issues found: fix, re-QA
Health Check
check-everything runs these in order — reason, align docs, record learnings, conform, test, tidy — pausing after each so you steer.
check-reasoning → qa-philosophy → find-learnings → qa-code → qa-tests → janitor
Harness health Check
qa-harness audits the harness as a system across the five pillars.
qa-harness → qa-skills → optimise-context → qa-philosophy
Grow the harness
Find a capability gap (or source a known skill), judge it with vet, reshape until it passes, then build it. vet rejects → revise → re-vet.
find-harness-improvements → vet → brainstorm → build-claude-skill
↩
source-a-skill→vet— alt start: a known skill↩
brainstorm→vet— reshaped: re-vet
Debug & Learn
Root-cause behind a failing test, then capture the lesson.
- ↩
debug→debug— 3 fails: question the architecture
Build the epic
Execute a whole epic, sprint by sprint; each ships its own PR to main. Hands-off auto-merges each gated sprint.
- ↩
build-sprint→build-epic— next sprint (loop until done)
Build the sprint
Front-load decisions, then ship a sprint's tasks — each its own gated PR into the sprint branch; hands-off auto-merges on green.
- ↩
build-sprint→build-sprint— next task (one PR each)