Fifteen reusable procedures for AI coding agents. From the first plan to independent proof, use them with the tools you already have.
Claude Code, Codex, Cursor, OpenCode, pi — and any tool that reads SKILL.md.
npx skills@latest add mneves75/skillsgit clone https://github.com/mneves75/skills.git ~/.claude/skills/mneves-skillsChoose skills and agents during installation.
Start with the work in front of you. Each guide includes triggers, prerequisites and an example.
Challenge assumptions. Check the evidence.
Audit application security and privacy using current OWASP guidance, plan remediation and apply authorized safe fixes. Adversarial review and independent acceptance distinguish source evidence from deployed controls.
Try: “audit security and privacy; fix confirmed local issues; do not deploy”
Added in 1.34.0Read the guideGate staging and production deploys on evidence: data safety, recovery, money, operations, billing and ownership. A ledger records each control, a verdict per target, and a fix plan for every gap.
Try: “is this ready for production? give me the fix plan”
Added in 1.37.0Read the guideStructured, isolated AI review of a fixed Git target. Ships its helper, security tests and its own engine defaults.
Try: “use autoreview on my local changes, including P3 findings”
Read the guideA fresh context checks frozen criteria against real evidence: tests, renders and the actual binary. PASS, FAIL or BLOCKED; only PASS allows “done.”
Try: “verify it”, “prove it”, “fix all”
Read the guideA bounded audit-and-ship pass: slop, performance, agent DX and PR triage. Falsifiable completion, existing authority and requested stops stay intact.
Try: “superaudit this repo, items 1 and 2”
Read the guideDecide, coordinate and hand over.
Objectives and incident reviews are guides and copy-ready prompts, not skills. Check each draft against your own evidence.
The main session owns decisions and acceptance. Route bounded, ordinary and difficult work by capability; codex-lane preserves multi-round context. Model bindings have one Defaults owner.
Try: “delegate”, “use codex”, “heavy task”
Read the guideStress-test a draft through first principles, relevant research, one combined challenge and distinct alternatives. Returns the improved deliverable.
Try: “challenge this”, “stress-test”, “best possible answer”
Read the guideA clipboard-ready handoff that opens a discussion. The receiver reviews first and decides whether the task is still worth doing. Portable anchors, no filesystem paths.
Try: “handoff the flaky login test”, “write a handoff for this bug”
Read the guideDeploy the pushed commit, prove it live, then tag. Staging preserves existing authority; production requires explicit approval of the exact candidate and target.
Try: “ship-staging”, “ship-prod”
Read the guideMake the artifact. Understand the project.
Build repeatable project-local Markdown exports as responsive HTML, editable Word and print-ready PDF, with shared styling and content/visual checks.
Try: “write a renderer for my Markdown reports”
Read the guideGenerate coherent iOS or Android screen and flow images, then inspect legibility, consistency and mobile conventions. Images only, never code.
Try: “design a 4-screen iOS onboarding flow”
Read the guideAudit and improve ChatGPT Search discovery with current official guidance. Keep search access, training choices, citations and referral measurement separate, with privacy and provenance intact.
Try: “improve this site's ChatGPT Search visibility”
Read the guideAssess how well a repo supports the agent's context, action, verification and iteration loop. Nine pillars, a percentage and maturity levels L1–L5. Runs locally.
Try: “why does the agent struggle here?”
Read the guideMake understanding stick.
Explain an idea with the Feynman method, calibrated to a child, layperson, executive, junior or expert. Answers in the listener's language.
Try: “explain OAuth to my dad”, “ELI5”, “explique de forma simples”
Read the guideYou explain the codebase; the agent reads the code, asks what you skipped and turns gaps into SM-2 flashcards in a per-project SQLite file.
Try: “let me explain the auth flow”, “quiz me”, “what's due?”
Read the guideA separate collection of 36 engineering workflow skills from Lauren Tan's pstack, through the pstack-claude port. Start with poteto-mode; it is not included in the default install.
The skills CLI links folders where your tools look. Use -g for user-wide installation or --all -y for no prompts. A plain Git clone works too.
Each SKILL.md begins with a description. Your agent loads the matching procedure, or you call it directly: /mneves-verify.
The steps run with your tools and permissions. Some skills ship helpers; others are Markdown only. A workflow preserves your scope and existing authority.
The readiness tooling runs locally. Image generation may send authorized prompts and references to your configured provider. Autoreview sends selected review inputs after credential scanning. Each procedure states its own requirements and limits. This site loads no third-party resources and sets no cookies.
mneves-agent-readiness includes a Bun + TypeScript assessor with no runtime dependencies. Point it at the project you want to assess.
cd /path/to/your-project
bun --bun /path/to/skills/tools/readiness-check.ts \
--format=html --output=report.html
# Static checks only
bun --bun /path/to/skills/tools/readiness-check.ts \
--skip-tests --skip-build
# Gate a CI job
bun --bun /path/to/skills/tools/readiness-check.ts \
--min-level=3Portable folders, clear triggers and verifiable behavior. The contributing guide covers the rules.
Read the contributing guidemneves-production-readiness: gates staging and production deploys on evidence-backed readiness.Object.prototype.mneves-fable-orchestrator routes building and reviewing to different models.