Domek-Labs
OrganizationSpec-driven agentic coding loops for Claude Code: every issue carries its spec, every repo its conventions, every loop a fixed, auditable 5-stage pipeline.
Categories
Indexed Skills (8)
github
Git and GitHub interaction conventions for loop commits, pushes, and PRs. Use gh / the GitHub API for writes when the local .git is read-only (constrained containers); set the commit author from AGENTS.md commit_identity; never a company/shared email; no Co-authored-by trailers unless configured (no_unconfigured_coauthors); PR body conventions (Closes #, standards block) and the squash-merge co-author caveat. Triggers: github, git push, gh cli, commit identity, commit author, coauthor, co-author, pull request, pr body, squash merge.
create-issue
Create a GitHub issue from a feature request or idea, with a full spec standard (Idea/Spec/AC/Files-To-Touch/Test-Plan/Out-of-Scope). Multi-repo. AGENTS.md in the repo overrides default templates. Multi-type system: --type=<code|research> flag plus auto-inference. Templates from references/issue-templates/<type>.md. Auto-injection of AGENTS.md ac_templates into the AC block plus 6-trigger issue-type detection (docs/epic/secret/mcp/live-ping/live-service). Codebase-memory auto-suggests files-to-touch and hotspots. Optional auto-handoff to /work-issue. Triggers: /create-issue, create issue, new issue, file issue, file feature, spec issue, idea as ticket.
init-agents
Create AGENTS.md (per-repo coding-standards spec) for a Git repo. YAML frontmatter with 11 mandatory v3 fields (branch_pattern, default_branch, pr_base, commit_format, syntax_check, smoke_test, deploy_command, linter, hard_gates, default_oos, ac_templates) plus the optional commit_identity field (12 total) + Markdown body (architecture, code conventions, test conventions, known gotchas). codebase-memory provides intelligent defaults from the repo cluster. Mandatory bootstrap before the first /work-issue run in a repo. Triggers: /init-agents, init agents, create agents.md, coding spec bootstrap, initialize repo standards.
loop
Loop-workflow umbrella. Routes to /init-agents (bootstrap), /create-issue (genesis), /plan-issues (wave planning), /work-issue (5-stage execution), /close-out (ordered merge), or /run-loop (full pipeline in one command). Triggers: "loop", "loop workflow", "work issue", "spec build", "coding loop", "github workflow", "github loop", "spec build loop". Routes when intent is unclear.
work-issue
Drive a GitHub issue through a 5-stage spec→build loop (Validator → Implementer → Tester → Critic → Closer). Multi-repo. Pure-reader — all standards values come from AGENTS.md at the repo root (no skill defaults). Multi-type system: reads the loop-type:<type> label and dispatches to a type-specific subagent brief (code|research). Mandatory pre-flight checks AGENTS.md existence, YAML parseability and completeness. Codebase-memory mandatory pre-flight. Auto-PR + merge on APPROVE. Triggers: /work-issue, work issue, loop workflow, drive issue, spec build loop.
close-out
Post-work lifecycle closer. Merges wave branches in wave order, auto-resolves additive conflicts, closes issues, and cleans up branches. No GitHub Projects required — tracks state exclusively via issue open/closed state and wave labels. Pipeline position: work-issue → close-out. Triggers: /close-out, close out, merge waves, ship waves, finalize waves, merge after work.
plan-issues
Wave-based execution planner. Assigns parallel-safety waves to open issues before the work-issue loops start. Prevents file collisions, migration-number clashes, and dependency-order violations in parallel runs. Outputs a wave plan for human review; no code is written. Pipeline position: create-issue → plan-issues → work-issue. Triggers: /plan-issues, plan issues, wave plan, plan before work, assign waves, parallel plan.
run-loop
Single-command pipeline entry point. Chains create-issue → plan-issues → work-issue with a mandatory human review gate between plan and work. Reduces a three-command sequence to one command plus one confirmation. Triggers: /run-loop, run loop, full loop, start loop, pipeline, run full pipeline.
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.