development-skills

End-to-end development workflow skills for Claude Code — from a GitHub issue to a merged PR.

Skills

Issue → Plan → Code

Skill	Description
research	Hypothesis-driven investigation under the four-state decision discipline (confirmed / rejected / inconclusive / deferred). Classifies hypotheses as empirical (subagent-verified) vs derivational (deduced in main context); posts the resulting plan to the issue body (umbrella-spawned sub-issue) or comment (single-scope issue). Depends on haru0416-dev/agent-skills (evidence-gated-review)
research-quaere	Quaere-based variant of research. Same workflow with the internal subagent contract delegated to quaere-evidence. Depends on haru0416-dev/quaere (quaere-evidence, quaere-semantic)
codex-plan-review	Review the implementation plan with Codex against the actual codebase before coding
implement	Execute a plan via the execution-loop discipline (Read → Plan → Execute → Review → Fix → Verify) with drift surfacing and done-check before completion. Reads the plan from the sub-issue body (umbrella-spawned) or comments (single-scope). Depends on haru0416-dev/agent-skills (execution-loop)
implement-quaere	Quaere-based variant of implement. Same workflow with the inner loop delegated to quaere-execution (Plan → Do → Study → Act). Depends on haru0416-dev/quaere (quaere-execution)
research-and-implement	End-to-end wrapper that runs research (Phase 1) then implement (Phase 2), with a branch baseline gate up front. Depends on haru0416-dev/agent-skills (evidence-gated-review + execution-loop)
research-and-implement-quaere	Quaere-based variant of research-and-implement. Sequences research-quaere then implement-quaere. Depends on haru0416-dev/quaere (quaere-evidence + quaere-execution)

Issue & PR drafting

Skill	Description
file-issue	Draft and file a GitHub issue following formatting conventions (semantic line breaks, LaTeX math, no local references). Includes an umbrella sub-issue variant (Parent: linkage, Goal/Scope/Out of scope/Acceptance shape) used by research when spawning sub-issues from an umbrella
file-adr	Draft an Architecture Decision Record (a timeless decision, distinct from an implementation schedule) and write the file under the project's ADR directory. Enforces classification against file-issue and a frozen-after-Acceptance discipline
file-pullreq	Draft and file a GitHub PR following the PR body skeleton, routed through the gh-post wrapper. Supports a gate mode that stops at user approval before posting
gh-body-conventions	Single source of truth for GitHub issue / PR body conventions — semantic line breaks, LaTeX-safe math, reference / exclusion policies, language defaults. Definition file referenced by the drafting and check skills, not a procedure
gh-body-check	Audit a drafted or filed GitHub issue / PR body against gh-body-conventions via a fresh-context subagent; any unresolved ⚠ blocks the caller

Documentation

Skill	Description
file-pubdoc	Draft README.md or visitor-facing markdown (top-level *.md, docs/**/*.md) from the canonical skeleton, complementing the public-doc-durability audit

Code → Review → Ship

Skill	Description
stage-commit-push	Stage, generate conventional commit message, commit, and push
codex-review	Run OpenAI Codex review with triage before PR creation
copilot-review	Create PR with GitHub Copilot review, poll for results, triage
review-pipeline	Orchestrator — runs the full flow from local changes to reviewed PR

End-to-end composite

Skill	Description
reimre	Full end-to-end wrapper — runs research-and-implement then review-pipeline back to back, with an automatic seam rule that skips the duplicate done-check at the boundary. Stops at the user-controlled merge gate inherited from review-pipeline.
reimre-quaere	Quaere-based variant of reimre. Runs research-and-implement-quaere then review-pipeline with the same seam rule and merge gate.

Quality Gates

Skill	Description
quality-list	Single source of truth for universal code-quality items, referenced by done-check and todo-check
todo-check	Preflight sweep of quality items before/during implementation
done-check	Post-hoc audit of quality items before declaring a task complete or requesting external review

Post-Ship Hygiene

Skill	Description
bug-to-contract	Promote review findings and bug fixes into runtime contract tests
finding-to-audit	Promote review findings into pre-commit audit rules that catch issues at diff-inspection time
gate-miss-to-issue	Promote a late-caught defect — one an earlier gate should have caught — into a development-skills issue proposing a fix to that gate's procedure
codex-contract-test-review	Narrow Codex pass to verify a newly added contract test actually expresses the claimed contract
driftreaper	Audit docstrings for drift against actual code behavior
breachreaper	Audit existing code for stock-detectable API-contract breaches

Language-specific

Skill	Description
languages/Rust/cargo-mutants	Configure and run cargo-mutants for Rust mutation testing
languages/Rust/rust-ffi-rule	Rules for implementing a Rust safe wrapper around an external (C / Fortran / FFI) call
languages/Cpp/stdlib-audit	Audit C++ source for known-bad standard library defaults via a TSV-driven, extensible rule table

Install

development-skills

claude plugin marketplace add ultimatile/development-skills
claude plugin install development-skills

gh-post

The skills that draft or post GitHub issue / PR bodies (file-issue, file-pullreq, copilot-review, review-pipeline, research, implement, and their wrappers / -quaere variants) route every body through the gh-post wrapper — a gh front-end that accepts bodies only via --body-file / --body-stdin and re-runs a hard-wrap validator before forwarding to gh. Install it and make sure gh-post is on PATH.

Quaere skills

The *-quaere variants (research-quaere, implement-quaere, research-and-implement-quaere, reimre-quaere) require the upstream Quaere skills (quaere-evidence, quaere-execution, quaere-semantic) installed under ~/.claude/skills/:

npx quaere-cli install --target claude

The original variants (without the -quaere suffix) instead depend on the archived haru0416-dev/agent-skills (evidence-gated-review, execution-loop).

Credits

The XML-block prompt structure used in codex-plan-review is inspired by the gpt-5-4-prompting skill from openai/codex-plugin-cc.