codex-goodies

Plugin codex-goodies in user538295/harness-goodies. Commands read from the repo's manifest — 2 days ago.

Skills (27)

• aaa

  Evaluate and improve ideas, product features, architecture, and code using evidence, industry best practices, research, gap analysis, and 3-4 world-class options.

• app-store-review

  Validate an app against every item of Apple's App Review Guidelines before App Store submission. Keeps a checklist file `AppReviewGuidelines.md` in sync with https://developer.apple.com/app-store/review/guidelines/, verifies each item against the app with parallel agents, and lists the items that ar

• bugfix

  Fix a bug end-to-end via a four-agent pipeline: reproduce it with a failing test (code inspection is NOT acceptable proof — only a failing test counts), fix it with TDD, run /iterative-review to clear every issue, then /commit. Use when the user reports a bug and wants it fixed, or invokes /bugfix <

• clean-code-review

  Structured clean code review — 132 checks across 7 groups (clarity, smells, solid, arch, tests, safety, ddd). Flexible targets — local changes (default), staged/unstaged/untracked, a git ref/range, a pull/merge-request link (GitHub/GitLab/Bitbucket), or explicit files (works without git). Runs scrip

• commit

  Write standardized Git commit messages in a direct, why-first style — Conventional Commits subject + a body of one or more `## H2` sections, each opening with a context paragraph then bullets. Use when the user asks to "write a commit message", "draft a commit message", "what should the commit messa

• create-tasks

  Generate the task-by-task breakdown for an existing team plan — a Mermaid dependency graph, vertical-slice phases, single-role checkbox tasks with per-task tests, and a mandatory close-out task — written to its own Markdown file. Use this whenever the user has a `<feature-name>-team-plan.md` (produc

• da-review

  Single-pass devil's advocate review. Finds flaws without auto-fixing. Use when the user invokes /da-review or asks for a devil's advocate review, or when another skill needs a read-only adversarial review. Use the iterative-review skill to also apply fixes.

• doc-voice

  Apply a direct, problem-first, proof-led voice when writing or revising user-facing documentation — READMEs, handouts, user manuals, landing-page copy, getting-started guides, marketing-adjacent prose. Use when a task involves drafting, rewriting, or critiquing copy meant to convince a developer to

• documentation-standard

  Comprehensive documentation management for software projects. Use when creating, updating, organizing, or refactoring project documentation, including architecture docs, ADRs, user manuals, and development guides. Covers structure, naming conventions, content standards, and maintenance workflows.

• double-check

  Config-driven, read-only review fan-out: runs every item listed in the project's `.double-check.json` — skills/commands, agents, MCP servers, CLI tools, Markdown rule files — on one target and merges all findings into one-line results. Targets: local changes (default), a commit, a commit range, file

• estimate

  Story Point estimation of a brief, plan, task list, epic or single task, calculated from planned hours at the scale's anchor or else placed by blind comparison with the project's delivered features, kept with their real effort in `.harness-goodies/estimations/`. Every run also collects what the proj

• feature-refinement

  Guide a rough feature idea into a well-scoped Feature Brief ready for /plan-maker or /plan-maker-for-team. Use when the user says "refine this feature", "help me scope this", "write a feature brief", "flesh out this idea", "is this the right feature to build", or invokes the feature-refinement skill

• implement

  Execute an existing plan or task-breakdown file — implement the next task, or loop through every remaining task — each task built test-first (TDD), self-reviewed, and committed as its own commit. Use when the user runs /implement or explicitly asks to implement/execute the tasks in a specific plan f

• iterative-review

  Spawn multiple devil's advocate agents in parallel — plus a Brooks-Lint reviewer when /brooks-review is available, and a one-off clean-code catalog pass in cycle 1 when /clean-code-review is available — to review, then spawn fix agents to address all moderate+ issues, re-run tests after each fix pas

• llm-wiki

  Set up or operate a generic LLM Wiki (Karpathy pattern) in any project, for any non-product domain — research, personal knowledge management, reading a book, course notes, competitive analysis, hobby deep-dives, learning. Use when the user asks to "set up an LLM wiki", "apply the Karpathy wiki patte

• md-reviewer

  Semantic and logic consistency reviewer for Markdown documentation. Use when reviewing documentation for terminology consistency, contradictions, cross-reference validity, formatting standards, voice/tone analysis, technical accuracy, and duplicate detection. Supports master-follower validation wher

• options

  Turn a decision or concern into a structured set of options with honest pros/cons and a firm recommendation. Use when the user asks "what are my options", "lay out the options", "pros and cons of X", "help me decide between", "weigh the tradeoffs", "which approach should I take", or otherwise wants

• plain-language

  Write for non-expert decision-makers — product owners, stakeholders, end users who don't know the codebase or the technology. Given something to explain ("explain this simply", "what does this error mean", "put this in plain language", "translate this for a stakeholder", "explain the tech so I can d

• plan-maker-for-team

  Generate a structured, role-split team development plan in Markdown from a feature brief. Use this whenever the user wants to turn a feature brief or a feature request into an actionable engineering plan that splits work across frontend, backend, and tester roles, with a layered architecture, Given/

• plan-maker

  Use this skill whenever the user asks to make a plan, create a plan, update a plan, or break down work into tasks. Triggers on phrases like "make a plan for", "create a plan", "plan this out", "write up a plan", "update the plan for", "add tasks to the plan", "how should we implement X", or any requ

• quick-plan

  Lightweight task decomposition — define a measurable goal, success criteria, and 4–12 actionable steps before doing any work. Use when the user says "plan this", "break this down", "decompose", "quick plan", "what are the steps", or when a task is complex enough to benefit from upfront structure. Ou

• session-log

  Manage session prompt logging and usage totals (on / off / status / usage)

• skill-packager

  Package a Claude Code skill folder as a ZIP file ready to upload to Claude Desktop or claude.ai via Settings → Customize → Skills → Create skill → Upload. Use when the user asks to "package this skill", "zip this skill for Desktop", "make the Desktop version", "bundle the skill for upload", "produce

• status-report

  Report status of a long-running background task. No args = one-shot. With interval (e.g. /status-report 10 mins) = recurring. /status-report off/cancel/stop = cancel the monitor.

• summarize

  Explain the findings and key lessons of supplied content (pasted text, URL or attached file) within a requested reading time, or by default in at most 24 lines of at most 255 characters each.

• vertical-slicer

  Decompose a feature into an ordered set of vertical slices — each delivering one user-observable behavior end-to-end, the first a walking skeleton — and never drift into horizontal or hybrid slicing. Use whenever the user wants to break a feature, story, epic, or initiative into increments, slice wo

• wrap-up

  Close out a work session: summarize what was done and why, run the tests and devil's advocate, then audit commit hygiene, acceptance criteria, and tracker status. Never changes anything (fix, commit, tracker) without an explicit yes, and blocks wrap-up on a failing test or devil's-advocate finding.

Subagents (1)

• devils-advocate

  Use this agent when you need to critically evaluate, stress-test, or challenge the output of another agent or reasoning process. Trigger this agent after any significant analysis, plan, recommendation, or code output has been produced to catch blind spots, false assumptions, hallucinations, logical

Hooks (1)

• SessionStart

  python3

Similar plugins