qnamy
UserAI-assisted development workflow showcasing cross-model planning, implementation, and review with Claude Code and Codex.
Categories
Indexed Skills (10)
comment-resolve
Verify whether an author's response to a review finding (a reply or a code change without a reply) actually resolves a thread I started, then recommend whether to resolve it and whether to revote. Follow pr-review for resolution criteria and unblocking conditions. The caller performs scanning, retrieval, resolution, voting, and platform API calls.
confluence-doc
Structure development documents such as designs, definitions, and proposals into readable Confluence pages, render Mermaid natively through a caller-injected diagram macro when one is available (otherwise degrade it to tables/ASCII/Expand), and publish through the Atlassian MCP. Load each document type's section skeleton from the matching file under skeletons/ on demand while using one type-independent engine for input collection, rendering, diagram handling, and publication. The caller injects publication targets (cloudId, space, parent, naming, accounts, status labels, and the optional Mermaid macro identity). Do not generate document content or judge its quality; upstream owns both.
deploy-approval
Review the changes targeted by a deployment approval request, decide whether to approve or hold based on deployment-blocking problems, and recommend advancing workflow state when the approval quorum is met. Use pr-review criteria (blocking means issue-level). The caller records approval signals, counts the quorum, and performs state transitions.
dev-full
Orchestrate large work such as new features, modules, and structural changes through the full lifecycle—planning (grounding + routing) → design → cross-model design review before code → approval gate → orchestrated execution → cross-model code review → final wave (Coverage, Quality, Runtime, Scope). Use symmetric cross-model review where design = Claude production → Codex review and development = Codex production → Claude review. Invoked by `/harnie:dev-full` or the `/harnie:dev` router. (Internal track value stays `plan`.)
dev-quick
Orchestrate small tasks such as incidents, localized bug fixes, and minor changes through a lightweight inline flow without skipping stage-by-stage cross-model review (design = Claude production → Codex review; development = Codex production → Claude review). Recommend the full track when new components, boundary/contract changes, or architecture decisions are required. Invoked by `/harnie:dev-quick` or the `/harnie:dev` router.
harness-digest
Analyze a finished harnie:dev run's execution state, review ledgers, and archived unit reviews to propose harness improvements — instruction prunes, granularity fixes, tier changes — with measured evidence. Propose only; the user chooses what to adopt, and incident-derived invariants are never auto-removed.
pr-delivery
Write profile-driven delivery content for completed changes: a PR title and selected body sections plus review-request content and review type. Use a caller-injected Delivery Profile for title convention, body sections, and review-request policy while leaving platform execution, branch, and merge strategy to the caller.
pr-review
Review submitted code changes in a PR at a senior-engineer standard, classify findings as `issue:`/`discuss:`/`nit:`, and recommend approval status. Judge only what is wrong, why it matters, and its severity; the caller determines execution details such as voting, mentions, comment placement, disclaimers, and platform operations. Use for human PR reviews and automated PR-review routines. In-loop development review is separate and lives in instructions/code-review.md.
dev
harnie's single development pipeline (0.11) — requirements grounding → (ARCH → review) → (task split + CONTRACT → review) → approval → build → cross-model review → verification, with stages skipped by size (S/M/L). Replaces dev-quick/dev-full. Invoked by `/harnie:dev`.
design-authoring
Author an architecture design or a detailed design as a standalone request (outside the /harnie:dev loops) by loading the canonical designer gates and the matching altitude output contract. Thin wrapper — it applies agents/harnie-designer.md and instructions/design-authoring-{arch,detail}.md by reference and restates neither. For reviewing an existing design, the criteria live in instructions/design-review.md instead.
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.