nine reusable skills, one Codex plugin, and 14 ready-to-run prompts.
use a skill for repeatable behavior across tasks. use a prompt for one focused workflow.
| format | use it when | invoke it with |
|---|---|---|
| skill | you need repeatable behavior across tasks | $skill-name |
| prompt | you need one focused workflow | /prompts:prompt-name |
install from the skills.sh package page:
npx skills add thoriqakbar0/skillsselect the skills and agents that you need. restart the agent after installation.
install one skill globally for Codex without prompts:
npx skills add thoriqakbar0/skills --skill coding-standards --agent codex --global --yesinvoke an installed skill by name:
$coding-standards review this TypeScript change and fix the verified defects
| skill | use it for |
|---|---|
coding-standards |
TypeScript design, implementation, review, and tests |
open-sourcing |
public-release preparation, licensing, secrets hygiene, CI, and packaging |
orchestrate |
substantial tasks with independent work for multiple agents |
pbt |
property-based testing that targets and proves plausible defects |
self-improve |
improvements supported by evidence from previous Codex sessions |
technical-documentation |
documentation that people and agents can use without hidden context |
unslop |
remove AI writing patterns while preserving meaning and voice |
visual-pr |
concise pull request descriptions with visual outlines for reviewers |
write-discoverable-code |
code names and messages that plain-text search can find |
- one workflow per skill: install only the behavior that you need.
- required evidence: Codex must run checks that match the risk.
- clear boundaries: each skill states its limits and approval points.
- plain-text instructions: inspect every instruction, script, and reference.
- standard format: each skill uses
SKILL.md.
coding-standards applies strict engineering rules to TypeScript changes.
use it for implementation, architecture, code review, refactoring, or test design. it checks the complete path from unknown input to observable behavior.
it guides Codex to:
- parse external data at system boundaries.
- make invalid states difficult to represent.
- model expected failures as typed values.
- keep domain logic separate from adapters and runtime code.
- give every side effect, promise, and resource a clear owner.
- make repeated write operations safe.
- prevent secrets from entering logs and errors.
- test behavior through real system boundaries.
- preserve strict TypeScript checks.
Codex must provide current evidence for each applicable rule. it must report any result that it cannot verify.
invoke it with $coding-standards.
open-sourcing prepares a private or existing public repository for a formal open-source release.
use it to:
- audit the complete Git history and related GitHub content for secrets.
- choose and apply a license consistently.
- close documentation, contributor, CI, and repository-setting gaps.
- set up versioning, release automation, and language-specific packaging.
- verify the repository from a new contributor's perspective before publication.
source: Trail of Bits' open-sourcing skill, licensed under CC BY-SA 4.0. see NOTICE.md for attribution and modification details.
invoke it with $open-sourcing.
orchestrate divides substantial work across agents while you keep control.
use it when a task has independent research, implementation, or review work. skip it when one agent can complete the task directly.
it guides Codex to:
- divide work into narrow assignments.
- prevent agents from editing the same scope.
- select a suitable reasoning level for each assignment.
- keep the main agent available to you.
- combine agent results into one verified outcome.
- keep approvals and important decisions with you.
invoke it with $orchestrate.
pbt builds property-based tests around likely defects and independent checks.
use it to design, implement, clean, migrate, refactor, or debug generated tests.
it guides Codex to:
- research risky behavior before proposing a property.
- show the test argument and wait for approval before editing.
- route Python to Hypothesis, supported languages to Hegel, browsers to Bombadil, and system faults to Antithesis.
- generate structured inputs that reach distinct states.
- prove each claimed defect with a regression or disposable mutation.
- clean related tests and make small testability refactors.
- preserve minimized failures and improve the generator around each bug family.
invoke it with $pbt.
self-improve turns repeated corrections into evidence-backed instruction improvements.
use it when the same correction or workflow problem appears across multiple sessions.
it can:
- list local Codex sessions.
- render a session as a readable transcript.
- find repeated corrections across sessions.
- audit installed skills for missing guidance.
- separate proposals for skills, project instructions, and global instructions.
- cite the sessions that support each proposal.
the skill proposes changes first. it does not edit instruction files without your explicit approval.
this skill requires the local Codex session database and rollout files. it does not work unchanged in other agent systems.
invoke it with $self-improve.
technical-documentation builds documentation that readers can follow and maintain.
use it for readmes, product documentation, contributor guides, agent instructions, or a full documentation review.
it guides Codex to:
- inventory documentation before writing.
- define the audience and reader outcome.
- preserve the current structure in an existing project.
- prefer durable wording for evergreen documentation.
- check commands, links, navigation, and referenced files.
- align
AGENTS.md,CONTRIBUTING.md, and compatible instruction files. - check multilingual parity when multiple languages exist.
- report validation results and remaining gaps.
this skill uses the AGPL-3.0-only license. see its SKILL.md for source metadata.
invoke it with $technical-documentation.
unslop removes common AI writing patterns without flattening the writer's voice.
use it for conversational output, documentation, articles, product copy, and other prose. it cuts puffery, vague claims, filler, repetitive structures, and awkward punctuation while preserving meaning and intended tone.
source: pstack by Lauren Tan, published under the MIT license.
invoke it with $unslop.
visual-pr creates or updates a pull request with a concise visual outline that helps reviewers understand why the change exists and how it is shaped.
use it for pull requests that need a compact view of changed data structures, file responsibilities, component trees, control flow, or data flow.
invoke it with $visual-pr.
source: HumanLayer's visual-pr skill, licensed under the MIT License. see NOTICE.md for attribution and modification details.
The poteto-mode plugin contains the complete pstack 0.14 Codex port from HustleCoding/pstack-codex. It includes 44 skills, 23 playbooks, 21 engineering principles, model routing, bounded collaboration, and real-artifact verification.
Add this repository as a Codex marketplace:
codex plugin marketplace add thoriqakbar0/skills
codex plugin add poteto-mode@thoriq-skillsStart a new task after installation, then invoke $poteto-mode. Run $setup-pstack once to write the recommended Codex model and runtime policy.
write-discoverable-code makes code easier for coding agents and humans to find with plain-text search.
read the source article: How coding agents read your code.
use it when writing or renaming functions, types, constants, files, error messages, or documentation comments.
it guides Codex to:
- use domain-specific names that produce unique search results.
- keep one spelling and one definition for each concept.
- put important constraints beside exported definitions.
- preserve complete event names, flags, and error prefixes as searchable literals.
- keep each file focused on one searchable concept.
- colocate tests with the behavior that they specify.
invoke it with $write-discoverable-code.
the prompts collection provides ready-to-run workflows for GitHub work, tests, planning, refactoring, and long tasks.
use a prompt when you need a repeatable command without installing a complete skill.
| prompt | use it to | what it does |
|---|---|---|
gh-address-pr-comments |
resolve selected PR review comments | applies focused changes and reports unresolved comments |
gh-commit |
divide local changes into focused commits | avoids unrelated files and never stages the complete worktree |
gh-fix-ci |
find and fix the first causal CI error | runs the smallest local check that covers the failure |
gh-review-pr |
review a PR for correctness, tests, and material risk | separates required fixes, suggestions, and questions |
gh-ship |
commit selected files, push a branch, and create a PR | stops on main or master and avoids git add . |
| prompt | use it to | what it does |
|---|---|---|
box-task |
send an improved execution brief to Codex in a new or named Ascii Box | preserves intent, adds acceptance checks, and records Playwright evidence for browser work |
codebase-audit |
audit a complete codebase for useful simplifications | inventories every subsystem, validates findings, and leaves the repository unchanged |
de-slop |
find AI artifacts and unrelated cleanup before a PR | shows findings first and changes only the selected items |
design-doc |
turn an implementation idea into a reviewable design | produces goals, design, alternatives, open questions, and an implementation plan |
html-artifact |
build a self-contained interactive HTML artifact | saves one browser-tested file with accessible controls and clear information hierarchy |
make-tests |
add behavior tests for the current change | covers success, boundary, failure, and regression paths |
work-forever |
continue long work with few questions | states assumptions, monitors long commands, and keeps destructive actions blocked |
box-task keeps the original request, then adds scope, assumptions, acceptance checks, and an evidence contract. Browser tasks record Playwright traces, videos, screenshots, logs, and a downloadable artifact manifest.
| prompt | use it to | what it does |
|---|---|---|
new-cmd |
turn a conversation or request into a reusable command | matches the active agent runtime and reports the exact invocation |
new-skill |
turn a conversation or request into a reusable skill | creates only the files and scripts that the workflow needs |
copy all prompts into Codex:
mkdir -p "$HOME/.codex/prompts"
rsync -a prompts/ "$HOME/.codex/prompts/"this command replaces prompt files with matching names. review local changes before you copy customized prompts.
invoke a copied prompt by name:
/prompts:gh-review-pr 123
use manual installation when you do not want the skills CLI.
git clone https://github.com/thoriqakbar0/skills.git
cd skills
mkdir -p "$HOME/.codex/skills/technical-documentation"
rsync -a technical-documentation/ "$HOME/.codex/skills/technical-documentation/"replace technical-documentation with the skill name that you want. copy the complete directory because a skill can use bundled resources.
pull the current repository version:
git pull --ff-onlyrun the applicable rsync command again. this updates matching files but does not remove obsolete files.
AGENTS.md contains Thoriq's shared agent policy and response style.
treat this file as personal configuration. review and merge its rules with your project instructions.
.
├── AGENTS.md shared agent instructions
├── README.md repository guide
├── coding-standards/ TypeScript engineering skill
├── open-sourcing/ public-release readiness skill and references
├── orchestrate/ multi-agent coordination skill
├── pbt/ property-based testing skill
├── prompts/ reusable Codex prompts
├── self-improve/ session review skill and script
├── technical-documentation/ documentation skill and references
├── unslop/ prose editing skill
└── write-discoverable-code/ code discoverability skill
review each skill before use. a skill can direct an agent to run tools with your permissions.