JUNERDD Skills

Reusable AI agent skills published from a single repository.

Current collection version: 0.2.3. Release notes are tracked in CHANGELOG.md and published through GitHub Releases.

This repository is a skill collection, not a single-skill package. Installable skills live under skills/, and each subfolder is meant to be independently installable and expanded over time. The root VERSION file tracks the published version of the collection as a whole using SemVer; Git tags and GitHub Releases use the vX.Y.Z form. Individual tools or subpackages may keep their own runtime versions when needed.

Stable production URL: https://junerdd-skills.vercel.app

Website (Vercel)

The companion landing page is a Next.js app in web/. English is served from unprefixed paths such as / and /skills/<slug>; Chinese is served from /zh-CN and /zh-CN/skills/<slug>.

In the Vercel dashboard, import this GitHub repository.
Under Configure Project, set Root Directory to web (critical for this monorepo layout).
Add Environment Variable on Production: NEXT_PUBLIC_SITE_URL = your production domain (example: https://skills.example.com, no trailing slash). This should match the primary Production domain configured under Project → Settings → Domains so Open Graph, canonical URLs, sitemap.xml, and robots.txt stay consistent.
After the first successful Production deployment, open your GitHub repo Settings → General and set Website to that same production URL so the repository “About” box links to the live site.

🧭 Skills At A Glance

If you are deciding what to install, start here:

comment-strategist - add high-value code comments without comment noise
exhaustive-code-slimmer - exhaustively reduce maintained code while preserving behavior
reduce-reinvention - find duplicated effort and guide reuse-first consolidation
find-local-skill - decompose requests, then find relevant local skills
git-commit - draft a Conventional Commit message from the staged diff
mr - use and maintain the Git MR/PR helper CLI
split-commits - split a mixed working tree into focused local commits
multitask-coordinator - coordinate scoped subagent work with isolation and ownership boundaries
plan-mode - plan complex or risky work before editing
debug - debug runtime issues with an evidence-first logging workflow
hack-review - review whether an implementation relies on brittle hack-like shortcuts
receiving-hack-review - consume a hack-review report and verify each finding before changing code
regression-review - review code changes for user-visible behavioral regressions
receiving-regression-review - consume a regression-review report and verify each finding before changing code

📦 Install

If you want an agent to install this repository for you without copying files, tell it:

Fetch and follow instructions from https://raw.githubusercontent.com/JUNERDD/skills/refs/heads/main/docs/INSTALL.md

If the agent already has this repository open locally, it can read ./docs/INSTALL.md directly instead of fetching the raw GitHub URL.

The CLI examples below intentionally use the latest skills tool version to avoid mismatches with older local installs.

List the skills currently published from this repository:

npx skills@latest add JUNERDD/skills --list

Install a specific skill:

npx skills@latest add JUNERDD/skills --skill <skill-name>

Install globally for Codex:

npx skills@latest add JUNERDD/skills --skill <skill-name> -g -a codex -y

Examples:

npx skills@latest add JUNERDD/skills --skill debug
npx skills@latest add JUNERDD/skills --skill git-commit
npx skills@latest add JUNERDD/skills --skill mr
npx skills@latest add JUNERDD/skills --skill split-commits
npx skills@latest add JUNERDD/skills --skill multitask-coordinator
npx skills@latest add JUNERDD/skills --skill plan-mode
npx skills@latest add JUNERDD/skills --skill comment-strategist
npx skills@latest add JUNERDD/skills --skill exhaustive-code-slimmer
npx skills@latest add JUNERDD/skills --skill reduce-reinvention
npx skills@latest add JUNERDD/skills --skill find-local-skill
npx skills@latest add JUNERDD/skills --skill hack-review
npx skills@latest add JUNERDD/skills --skill receiving-hack-review
npx skills@latest add JUNERDD/skills --skill regression-review
npx skills@latest add JUNERDD/skills --skill receiving-regression-review

Manual symlink install still works if you prefer not to use the agent prompt:

mkdir -p ~/.agents/skills
ln -s "$PWD/skills" ~/.agents/skills/junerdd-skill

🧱 Repository Model

The collection version lives in the root VERSION file.
Release notes live in CHANGELOG.md, and published GitHub releases should use matching vX.Y.Z tags.
Use the release workflow to prepare a collection release. It opens a release PR that updates VERSION, the README version line, website version metadata, and CHANGELOG.md from one workflow input; merging that PR publishes the matching GitHub Release.
Each installable skill lives under skills/<skill-name>/.
Each skill owns its own SKILL.md plus any optional agents/, references/, scripts/, or assets/ directories.
Root-level files describe the repository as a collection. Skill-specific behavior and deep operational details stay inside the relevant skill folder.
Shared repository assets such as screenshots can live outside skills/ when they are not part of the installable package itself.

🗂️ Current Skills

Use the anchor list above for a quick jump, then read the section that matches your task.

`comment-strategist`

skills/comment-strategist/ is for documenting existing code without adding low-value comment noise. It focuses on intent, contracts, constraints, field meaning, and control-flow rationale instead of rewriting syntax in prose.

Install:

npx skills@latest add JUNERDD/skills --skill comment-strategist

Best for:

documenting exported functions, interfaces, classes, and config objects
replacing outdated or redundant comments with durable explanations
adding guided comments inside complex logic while preserving the local comment style

Key entry points:

Workflow and guardrails: skills/comment-strategist/SKILL.md
Optional runtime metadata: skills/comment-strategist/agents/openai.yaml

`exhaustive-code-slimmer`

skills/exhaustive-code-slimmer/ exhaustively searches for behavior-preserving code reductions. It combines audit scripts, deletion-first candidate search, oracle design, and an approval gate for architecture-level refactors so slimming improves maintainability instead of producing dense or risky code.

Install:

npx skills@latest add JUNERDD/skills --skill exhaustive-code-slimmer

Best for:

finding removable files, branches, exports, dependencies, wrappers, and duplicate logic
running deletion candidates against a build/test/lint/smoke oracle before accepting changes
identifying architecture problems that block safe code reduction and presenting DX-oriented options before refactoring

Key entry points:

Workflow and guardrails: skills/exhaustive-code-slimmer/SKILL.md
Code-slim audit script: skills/exhaustive-code-slimmer/scripts/code_slim_audit.py
Exhaustive shrink script: skills/exhaustive-code-slimmer/scripts/exhaustive_shrink.py
Optional runtime metadata: skills/exhaustive-code-slimmer/agents/openai.yaml

`reduce-reinvention`

skills/reduce-reinvention/ identifies and reduces duplicated effort across code, libraries, services, templates, docs, platform workflows, and architecture decisions. It combines a reuse-first workflow with audit scripts and decision templates so teams can adopt, adapt, consolidate, or justify divergence with evidence.

Install:

npx skills@latest add JUNERDD/skills --skill reduce-reinvention

Best for:

auditing duplicated implementations and overlapping reusable assets
deciding build-vs-reuse/buy with ownership, maintenance, security, and migration tradeoffs
creating reuse catalogs, ADR/RFC records, migration plans, or golden-path guidance

Key entry points:

Workflow and guardrails: skills/reduce-reinvention/SKILL.md
Reuse playbook: skills/reduce-reinvention/references/reuse-playbook.md
Audit checklist: skills/reduce-reinvention/references/audit-checklist.md
Decision matrix: skills/reduce-reinvention/references/decision-matrix.md
Output templates: skills/reduce-reinvention/references/templates.md
Reinvention audit script: skills/reduce-reinvention/scripts/reinvention_audit.py
Reuse catalog script: skills/reduce-reinvention/scripts/reuse_catalog.py
Optional runtime metadata: skills/reduce-reinvention/agents/openai.yaml

`find-local-skill`

skills/find-local-skill/ helps agents decompose a request into deliverables, workflow phases, tools, domains, and implicit prerequisites before inspecting available local skills, selecting the ones that match, and applying those workflows before normal analysis. It includes a local scanner for plain project skills/ folders, Cursor, Claude Code, OpenCode, Codex, shared Agent Skills roots, and plugin skill caches.

Install:

npx skills@latest add JUNERDD/skills --skill find-local-skill

Best for:

finding relevant local skills before planning or implementation
routing ambiguous or multi-phase requests through explicit skill selection
surfacing implicit prerequisite skills before deeper requirement analysis
auditing available skill coverage across plain project skills/, Cursor, Claude Code, OpenCode, Codex, and shared Agent Skills roots
distinguishing plugin skills with names such as product-design:index

Key entry points:

Workflow and guardrails: skills/find-local-skill/SKILL.md
Local skill scanner: skills/find-local-skill/scripts/list_agent_skills.py
Optional runtime metadata: skills/find-local-skill/agents/openai.yaml

`git-commit`

skills/git-commit/ drafts a Conventional Commit message from the staged diff only. It is intentionally narrow: it reads what is already staged, proposes one accurate message, and does not mutate Git state.

Install:

npx skills@latest add JUNERDD/skills --skill git-commit

Best for:

generating a clean commit subject from the current index
checking whether a staged batch is too mixed for one honest commit message
keeping commit wording grounded in staged files instead of unstaged work

Key entry points:

Workflow and guardrails: skills/git-commit/SKILL.md
Optional runtime metadata: skills/git-commit/agents/openai.yaml

`mr`

skills/mr/ supports the mr Node CLI for generic Git MR/PR workflows. It covers target aliases, MR branch strategies, default detached mode, request providers or custom request commands, configuration, conflict resume, install/update/uninstall behavior, automatic update notices, and maintenance of the TypeScript CLI implementation behind the tool.

Install:

npx skills@latest add JUNERDD/skills --skill mr

Best for:

creating or previewing Git merge requests or pull requests with mr, mrm, mrt, or mrp
checking for a missing local mr install and installing it after user confirmation
choosing between merge, rebase, merge-target, pr, and default detached-mode flows
configuring CNB/GitHub/GitLab providers or a custom mr.requestCommand
understanding non-blocking update notices and the environment variables that disable them
handling stopped merge/rebase states by preserving CLI-owned resume paths
maintaining the TypeScript/Pastel/Ink/Zod implementation behind the CLI

Key entry points:

Workflow and guardrails: skills/mr/SKILL.md
CLI reference: skills/mr/references/mr-cli-reference.md
Optional runtime metadata: skills/mr/agents/openai.yaml

`split-commits`

skills/split-commits/ helps break a mixed working tree into a sequence of focused local commits. It stages one logical batch at a time, asks $git-commit for a message, and requires explicit confirmation before each git commit.

Install:

npx skills@latest add JUNERDD/skills --skill split-commits

Best for:

separating unrelated concerns in the same working tree
isolating refactors from behavior changes
building a short series of reviewable local commits without pushing

Key entry points:

Workflow and guardrails: skills/split-commits/SKILL.md
Optional runtime metadata: skills/split-commits/agents/openai.yaml

`multitask-coordinator`

skills/multitask-coordinator/ coordinates non-trivial multi-step work where an agent may use background subagents or local task decomposition. It keeps the parent agent responsible for framing, decomposition, shared contracts, delegation decisions, worker ownership boundaries, isolation choices, synthesis, verification, and user communication.

Install:

npx skills@latest add JUNERDD/skills --skill multitask-coordinator

Best for:

deciding whether complex repo work should stay local, be decomposed, or be delegated
assigning disjoint worker scopes in large repositories, monorepos, multi-root workspaces, dirty worktrees, or isolated worktrees and branches
keeping shared contracts, package exports, sequencing, and destructive migration boundaries under parent-agent ownership
coordinating queued independent requests, async exploration, implementation, review, or verification without duplicating work
preserving parent-agent ownership of final integration, validation, and user-facing status

Key entry points:

Workflow and guardrails: skills/multitask-coordinator/SKILL.md
Optional runtime metadata: skills/multitask-coordinator/agents/openai.yaml

`plan-mode`

skills/plan-mode/ mirrors Cursor's Plan Mode loop for complex, ambiguous, risky, or multi-file work: create a disk-backed editable Markdown plan, research the codebase into file/code references, ask focused clarification questions, maintain buildable todos, pressure-test non-trivial plans with $grill-me, and build only after the plan is approved.

Install:

npx skills@latest add JUNERDD/skills --skill plan-mode

Best for:

planning implementation for broad or multi-file tasks before editing
tracing routes, data flow, architecture constraints, tradeoffs, or risky operations
maintaining an editable plan document with file references and checkbox todos
invoking $grill-me to pressure-test meaningful assumptions, risks, and rollout edges
building all or selected todos only after the user approves the plan

Key entry points:

Workflow and guardrails: skills/plan-mode/SKILL.md
Plan artifact helper: skills/plan-mode/scripts/plan_artifact.py
Architecture reference: skills/plan-mode/references/architecture.md
Optional runtime metadata: skills/plan-mode/agents/openai.yaml

`debug`

skills/debug/ provides evidence-first runtime debugging for application bugs, regressions, flaky behavior, and unclear runtime failures. It is the most operational skill in the repository and includes both workflow guidance and a local log collector.

Install:

npx skills@latest add JUNERDD/skills --skill debug

Key entry points:

Workflow and guardrails: skills/debug/SKILL.md
Operator reference: skills/debug/references/runtime-debugging.md
Local NDJSON collector: skills/debug/scripts/local_log_collector/
Optional runtime metadata: skills/debug/agents/openai.yaml

`debug` Skill Snapshot

The debug skill is designed to prevent speculative fixes by forcing a prove-it loop:

Generate precise hypotheses.
Attach to or start an authoritative logging session.
Add minimal temporary instrumentation.
Reproduce the issue and read the recorded log file.
Mark each hypothesis as CONFIRMED, REJECTED, or INCONCLUSIVE.
Apply a fix only after the root cause is proven.
Verify with fresh post-fix logs before removing instrumentation.

This keeps the skill focused on evidence, not guesswork.

`debug` Architecture

flowchart LR
  User["Developer / Operator"] --> Agent["Agent Runtime"]
  Agent --> Skill["debug/SKILL.md<br/>workflow + guardrails"]
  Agent --> Ref["runtime-debugging.md<br/>bootstrap + log format"]
  Agent --> App["Target app under debug"]
  Agent --> Logs["Temporary instrumentation"]
  Logs --> Collector["Local NDJSON collector<br/>same-origin dashboard + APIs"]
  App --> Logs
  Collector --> File["Session log file"]
  Collector --> UI["Live dashboard"]
  File --> Agent
  UI --> Agent
  Agent --> Fix["Proven fix + post-fix verification"]

`debug` Highlights

Evidence-first debugging instead of inspection-only reasoning
Minimal instrumentation with explicit cleanup after verification
Per-hypothesis logging and before/after comparison
Local collector bootstrap when the host does not already provide logging
Browser-first log transport for frontend debugging, with explicit prohibition on app-local proxy routes unless direct delivery is proven blocked

`debug` Runtime Support

The current debug skill is intentionally portable. It works with:

OpenAI Codex and similar local-skill runtimes
Agent shells that read ~/.agents/skills/<name>/SKILL.md
Custom agent frameworks that mount a skill folder and inject SKILL.md into context
Internal toolchains that want the collector, references, or workflow as reusable assets

If your runtime ignores skills/debug/agents/openai.yaml, the core logic is still fully available through skills/debug/SKILL.md.

`debug` Dashboard Preview

`debug` Collector

The bundled collector is a zero-dependency Python app built on the standard library. It accepts JSON log events, appends them to an NDJSON file, and serves a same-origin dashboard for live inspection.

For frontend and browser debugging, the intended transport is direct client-to-collector HTTP posting. The collector already handles CORS and preflight, so the skill should not create temporary Next.js API routes or other app-local proxy layers unless direct browser delivery has been proven blocked in the current host.

Collector endpoints:

POST /ingest
GET /health
GET /api/state
GET /api/logs
GET /api/logs/detail
POST /api/clear
POST /api/shutdown

Minimal smoke test:

mkdir -p .debug-logs
python3 skills/debug/scripts/local_log_collector/main.py \
  --log-file "$PWD/.debug-logs/demo.ndjson" \
  --ready-file "$PWD/.debug-logs/demo.json" \
  --session-id "demo-session"

`hack-review`

skills/hack-review/ reviews whether an implementation is relying on hack-like tactics instead of sound ownership and abstraction boundaries. It produces a coverage-led reviewer report that enumerates all distinct hack-risk findings discovered within scope, records intentional exceptions, and marks uncovered ownership boundaries explicitly.

Install:

npx skills@latest add JUNERDD/skills --skill hack-review

Best for:

identifying impossible-state fallbacks that hide a broken invariant
flagging symptom-masking patches that do not fix the root cause
catching duplicate abstractions or parallel wheels when a stable boundary already exists
showing which touched ownership boundaries were reviewed, skipped, or left uncovered

Key entry points:

Workflow and guardrails: skills/hack-review/SKILL.md
Report template: skills/hack-review/references/report-template.md
Optional runtime metadata: skills/hack-review/agents/openai.yaml

`receiving-hack-review`

skills/receiving-hack-review/ consumes a hack-review report and builds a disposition ledger for every finding, intentional exception, and open ownership coverage gap before deciding whether to fix, challenge, confirm, or carry it forward.

Install:

npx skills@latest add JUNERDD/skills --skill receiving-hack-review

Best for:

re-checking that a hack report matches the exact requested review scope
fixing ownership problems without mechanically deleting necessary guards
narrowing or disproving hack findings with stronger code-path evidence
closing or explicitly carrying forward Not covered ownership boundaries from the coverage ledger

Key entry points:

Workflow and guardrails: skills/receiving-hack-review/SKILL.md
Optional runtime metadata: skills/receiving-hack-review/agents/openai.yaml

`regression-review`

skills/regression-review/ reviews whether the current change set introduces user-visible behavioral regressions. It writes a coverage-led reviewer report that enumerates all distinct findings discovered within scope, records intentional visible changes, builds scoped behavior-graph deltas for affected surfaces when useful, and marks uncovered surfaces explicitly.

Install:

npx skills@latest add JUNERDD/skills --skill regression-review

Best for:

checking whether a refactor or feature work breaks user-facing flows
auditing changed defaults, loading states, retries, ordering, or exported output
tracing changed inputs, guards, transforms, and outputs through scoped behavior graphs
writing a review artifact that keeps severity aligned with the strongest unresolved finding while showing full reviewed coverage

Key entry points:

Workflow and guardrails: skills/regression-review/SKILL.md
Report template: skills/regression-review/references/report-template.md
Optional runtime metadata: skills/regression-review/agents/openai.yaml

`receiving-regression-review`

skills/receiving-regression-review/ consumes a regression-review report and builds a disposition ledger for every finding, behavior-graph delta, intentional visible change, and open coverage gap before deciding whether to fix, challenge, confirm, or carry it forward.

Install:

npx skills@latest add JUNERDD/skills --skill receiving-regression-review

Best for:

re-checking a regression gate against the current diff and baseline
reconciling behavior-graph deltas with findings and coverage rows
fixing only proven user-visible regressions instead of blindly following review comments
separating real regressions from intentional product deltas with stronger evidence
closing or explicitly carrying forward Not covered user-visible surfaces from the coverage ledger

Key entry points:

Workflow and guardrails: skills/receiving-regression-review/SKILL.md
Optional runtime metadata: skills/receiving-regression-review/agents/openai.yaml

🌱 Growing The Repository

When you add more skills later:

Create a new folder under skills/<skill-name>/.
Keep each skill self-contained so it can be installed independently.
Add or update agents/, references/, scripts/, and assets/ only when they materially help that specific skill.
Update the Current Skills section in this README with a one-line summary and relevant links.
Keep repo-level README content about the collection itself; move deep procedural detail into the skill that owns it.

🏗️ Repository Layout

.
├── LICENSE
├── README.md
├── docs/
│   └── images/
│       └── dashboard-overview.png
└── skills/
    ├── comment-strategist/
    │   ├── SKILL.md
    │   └── agents/
    │       └── openai.yaml
    ├── debug/
    │   ├── SKILL.md
    │   ├── agents/
    │   │   └── openai.yaml
    │   ├── references/
    │   │   └── runtime-debugging.md
    │   └── scripts/
    │       └── local_log_collector/
    │           ├── main.py
    │           ├── collector_server.py
    │           ├── collector_state.py
    │           ├── collector_browser.py
    │           └── static/
    ├── exhaustive-code-slimmer/
    │   ├── SKILL.md
    │   ├── agents/
    │   │   └── openai.yaml
    │   ├── references/
    │   │   ├── code_cleanliness_guide.md
    │   │   ├── dx_architecture_gate.md
    │   │   ├── language_tactics.md
    │   │   ├── oracle_design.md
    │   │   ├── research_basis.md
    │   │   └── transformation_catalog.md
    │   └── scripts/
    │       ├── architecture_dx_scan.py
    │       ├── code_slim_audit.py
    │       └── exhaustive_shrink.py
    ├── find-local-skill/
    │   ├── SKILL.md
    │   ├── agents/
    │   │   └── openai.yaml
    │   └── scripts/
    │       └── list_agent_skills.py
    ├── git-commit/
    │   ├── SKILL.md
    │   └── agents/
    │       └── openai.yaml
    ├── hack-review/
    │   ├── SKILL.md
    │   ├── agents/
    │   │   └── openai.yaml
    │   └── references/
    │       └── report-template.md
    ├── mr/
    │   ├── SKILL.md
    │   ├── agents/
    │   │   └── openai.yaml
    │   └── references/
    │       └── mr-cli-reference.md
    ├── multitask-coordinator/
    │   ├── SKILL.md
    │   └── agents/
    │       └── openai.yaml
    ├── receiving-hack-review/
    │   ├── SKILL.md
    │   └── agents/
    │       └── openai.yaml
    ├── receiving-regression-review/
    │   ├── SKILL.md
    │   └── agents/
    │       └── openai.yaml
    ├── regression-review/
    │   ├── SKILL.md
    │   ├── agents/
    │   │   └── openai.yaml
    │   └── references/
    │       └── report-template.md
    └── split-commits/
        ├── SKILL.md
        └── agents/
            └── openai.yaml

⚖️ License

Released under the MIT License.

Name		Name	Last commit message	Last commit date
Latest commit History 118 Commits
.github/workflows		.github/workflows
docs		docs
scripts		scripts
skills		skills
web		web
.gitignore		.gitignore
AGENTS.md		AGENTS.md
CHANGELOG.md		CHANGELOG.md
LICENSE		LICENSE
README.md		README.md
VERSION		VERSION

Folders and files

Latest commit

History

Repository files navigation

JUNERDD Skills

Website (Vercel)

🧭 Skills At A Glance

📦 Install

🧱 Repository Model

🗂️ Current Skills

comment-strategist

exhaustive-code-slimmer

reduce-reinvention

find-local-skill

git-commit

mr

split-commits

multitask-coordinator

plan-mode

debug

debug Skill Snapshot

debug Architecture

debug Highlights

debug Runtime Support

debug Dashboard Preview

debug Collector

hack-review

receiving-hack-review

regression-review

receiving-regression-review

🌱 Growing The Repository

🏗️ Repository Layout

⚖️ License

About

Topics

Resources

License

Uh oh!

Stars

Watchers

Forks

Releases 14

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

`comment-strategist`

`exhaustive-code-slimmer`

`reduce-reinvention`

`find-local-skill`

`git-commit`

`mr`

`split-commits`

`multitask-coordinator`

`plan-mode`

`debug`

`debug` Skill Snapshot

`debug` Architecture

`debug` Highlights

`debug` Runtime Support

`debug` Dashboard Preview

`debug` Collector

`hack-review`

`receiving-hack-review`

`regression-review`

`receiving-regression-review`

Packages