Use the Ralph skills¶
Drive adepthood development with the repo's agent skills, from a phone
session or the autonomous loop. Verified against adepthood CLAUDE.md,
.claude/skills/, and .claude/commands/ralph-tick.md as of 2026-07-31.
Prerequisites¶
- A Claude Code session in the adepthood repo (the skills live in
.claude/skills/; the orchestrator command in.claude/commands/) - For anything that commits: the
.venvactivated and gates runnable (quality gates)
The day-to-day skills¶
/continue-epic— pick up the next issue from the roadmap (prompts/github-issues/) and drive it to PR./triage-and-plan— analyze the codebase and generate a new epic of issues./preflight— run pre-commit, fix all failures, iterate until green./review-diff— self-review the current branch diff before PR./flare— turn a short bug/idea description into a grounded, Ralph-ready GitHub issue in one shot./graph— graph-first codebase questions (query the knowledge graph).
Supporting skills the workflow leans on: stay-green (the TDD loop),
ci-debugging (CI failures), address-feedback (iterate on the reviewer's
Verdict: comment), await-claude-review (wait for the verdict without
polling), and de-slopify / backlog-grooming for maintenance.
The autonomous loop¶
Start the fleet orchestrator:
/loop /ralph-tick
One re-entrant session runs a pool of up to 4 worktree lanes
(scripts/ralph/fleet.sh), each moving independently through the four
gates — TDD → local check-all → CI → review — with the orchestrator alone
merging (scripts/ralph/FLEET.md;
ADR 0007). Workers follow
the per-issue contract in scripts/ralph/PROMPT.md: one issue, one PR,
never merge, never touch main.
Verify¶
- A skill run ends with the state it promises (e.g.
/preflightends withpre-commit run --all-filesexiting 0;/continue-epicends with an open PR). - Fleet health: merged PRs post a Discord recap (
scripts/ralph/RECAP.md), andscripts/ralph/pick-next.sh/pr-ready.share the ground truth for what runs next and what may merge.