arm-pz/caffeinity

Produces tailored job-application material from a resume and a job description: a decoded JD table with must-have versus nice-to-have, resume bullets rewritten as action-task-result-impact, a keyword-gap placement map, five drafted resume summaries, a full LinkedIn rewrite, and twenty likely interview questions with sample answers. Use whenever the user wants the better version written: a resume rewritten for a role, stronger bullets, missing keywords worked in, a summary or LinkedIn headline and About section, or interview answers. Not for judging (scans, critiques, ATS checks, rejection reasons, gap analysis, LinkedIn audits, predicting questions): use the recruiter-lens skill. Not for running the search itself (finding roles, the application tracker, follow-ups, interview briefings): use the application-pilot skill. Optionally uses the third-party dani-aisystems recruiter-lens MCP server for checkable GitHub facts; it runs third-party code. Works standalone or with either companion.

Runs the job search around the applications: turns role criteria into a live shortlist, scores each role for fit with evidence, keeps a reviewable tracker (status, submission and follow-up dates, dedupe), flags roles gone quiet past an aging threshold, drafts follow-ups and interview briefings, and reviews the funnel weekly. Use when the user wants to find, organize, compare, track or follow up on roles: "find backend roles posted this week", "build my application tracker", "brief me before Thursday's interview". Not for writing application material (resume, bullets, summaries, LinkedIn, answer scripts): use applicant-kit. Not for judging (scans, ATS checks, critique, rejection reasons, gap analysis): use recruiter-lens. Markdown and CSV only: no CLI, no cloud state, no telemetry, no browser automation, no submit path; the user approves every application. If job-application-agent is installed and the user wants automated filling, batch submission, or a programmatic ledger, that skill owns it.

Picks which caffeinity skills to run, in what order, for a goal that spans more than one of them, and says what each step hands the next: build and publish a skill, get a project found, pass a release gate, grow organic search traffic, run a job search, make a decision, learn a skill, run cold outbound, launch a one-person product, or set up and keep the skills themselves healthy. Checks which of those skills are installed and names the missing ones instead of improvising their work. Use whenever someone asks which skill to use, what to run first, how the skills fit together, or describes a multi-step outcome such as "take this idea to a published skill" or "get my site ready to ship and ranking", even if they never name a skill. Not for a request one skill already owns (load that skill directly), and not for doing a step's work itself: the named skill does it.

Career positioning and the moves around it: personal brand and function-level visibility, promotion and performance-review narratives, executive presence, salary negotiation scripts, career pivots, networking that reaches opportunities before they are posted, interview preparation for the rooms you are already in, and a plan with immediate actions and ninety-day milestones. Use whenever someone wants to move up or across: how do I get promoted, negotiate a raise, am I ready for senior, pivot out of this role, prep for a promotion panel, people do not notice my work, even if they never say career plan. Not for the application documents themselves: resume and CV judging (use recruiter-lens) or tailored application material and LinkedIn rewrites (use applicant-kit).

Gate a cold-email lead list before it loads: sample ten rows, check company, title, size, and mailbox against the ICP, compare data providers on the same criteria, and return a GO or HOLD decision with the fail pattern and the re-run condition. Use when someone wants a list QA'd, checked, sampled, or approved before a send, asks whether a list is clean or safe to load, or is choosing between data providers for one filter set. Not for buying, scraping, or enriching contacts (out of scope), writing the email copy (wordsmith), or diagnosing a list that already sent (cold-email-ops).

Plan and audit a cold-email program at volume and keep it from degrading: computes inbox and domain capacity from a monthly send target, sets per-domain reply-rate floors, burn-replacement cadence, shadow-ban tests, and automation error budgets, and reports prospects-to-lead with per-client averages. Use when someone is scaling outbound, asks how many inboxes or domains they need, why reply rates are falling or a domain might be shadow banned, or reviews an AI outbound stack for failure points. Not for pre-load list QA (cold-email-list-gate), writing or polishing the emails themselves (wordsmith), inbound SEO (seo-audit, programmatic-seo), WhatsApp/RCS messaging (integrate-whatsapp, sent-two-way-messaging), or finding the market/ICP.

Designs a course for a group of learners (a class, cohort, workshop series, bootcamp or internal training): measurable outcomes, a module sequence with prerequisites, activities and assessments aligned to each outcome, a time budget that fits the hours available, and a rubric, then checks the plan with a script so no outcome goes untaught or unassessed. Use when someone is teaching others and asks for a curriculum, syllabus, course outline, module plan, cohort or workshop structure, or wants an existing course checked for gaps and overload, even if they never say "curriculum". Not for one person's own learning plan (use study-system), for being taught the topic (use topic-primer), or for marketing the course (use sales-copy).

Crash-tests a decision before you build it. Four sub-agents (Pragmatist, Skeptic, Specialist, Contrarian) draft the same task in separate contexts, attack each other's drafts anonymously, and a chair merges what survived, logs unresolved dissent, and checks the winner's claims about existing files against the source before adopting it. Use when the user says "crash test this", "second opinion", "council", "review panel", "pressure-test this", "I don't trust this answer", or faces a costly decision (architecture, plan, positioning, debugging hypothesis). Lighter plan-debate mode (Skeptic vs Contrarian, ~5 calls) for a verdict before committing. Optional Operator seat reviews cost, rollout and reversibility (+1 call). Custom role files and per-member models supported. About 9 calls, no API keys, no server. Not for factual lookups, small edits, debugging an actual crash, or reviewing code for defects: it examines a decision, not a diff.

Raw numbers turned into a decision: dataset interpretation, statistics with the business implication of each finding, chart specifications, pattern and anomaly detection, forecasts built on stated assumptions, segmentation, A/B test significance, dashboard metric selection and report templates, output as an executive summary, findings ranked by impact and the specific actions they support. Use whenever someone brings data, a spreadsheet, a CSV, an analytics export, an A/B result or an unexpected number and asks what it means, what to chart, whether the test won or which metrics to track, even if they never say analysis. Not for writing the pandas or R code that crunches the file (use production-code), for market research outside the user's own data (use research-brief) or for the strategic roadmap that follows (use strategy-memo).

Make 100 versions of the assistant fight to the death over one task. Spins up N sub-agents (default 100, --quick for 16), gives every one the exact same task plus a different strategy card (reasoning mode, workflow, strategy), then runs a single-elimination bracket: they attack each other's solutions, defend and revise, and a judge scores every match on a written rubric until one solution survives. Use when the user is not satisfied with an answer, calls it a bad answer, says try again or do better, says "arena", or asks to make them compete. Not for stress-testing a decision or plan before building it (use crash-test).

Create a new agent skill from a plain idea, with no prompt list to work from. Confirms intent before writing anything, decides what belongs in scripts versus references versus assets, writes a description that actually triggers, validates with tune-skill's check_skill.py, then hands off for boundaries and packaging. Use whenever the user wants to build, scaffold, author, start or design a new skill, describes a capability they want an agent to have repeatedly, points at an empty skill folder, or says "make me a skill for X", even if they never mention frontmatter or file layout. Not for turning a batch of existing prompts into a skill (use prompts-to-skill), tightening a skill that already exists (use tune-skill), or releasing a finished one (use prompts-to-skill-release).

Sets up discovery for a GitHub repo, agent skill, Claude Code plugin or MCP server so people and agents can find it by what it does, not only by its name: audits current visibility and same-name collisions, derives the queries people actually search, drafts the GitHub About text, topics, README opening, release notes and a skill page for a website, lists directories to submit to, and plans a re-check. Use whenever the user wants more stars, installs, traffic or visibility for a repo, skill or skill page, asks why a similar project shows up instead of theirs, or is about to publish a skill, plugin or MCP server. Not for finding repos or skills for a task, which is the reverse problem. Not for authoring a new skill (use draft-skill) or tuning a SKILL.md description so an agent triggers it better (use tune-skill).

Audit content for discoverability gaps with the Findable skill, then generate improved alternatives with Wordsmith. Use when the user wants to improve SEO, search visibility, or findability of a repo README, skill page, landing page, article, or marketing copy — and wants both the diagnosis and concrete rewrites. Chains findable (audit) → wordsmith (alternatives) in one workflow. Not for tuning SKILL.md trigger language (use tune-skill), not for code review, and not for resume optimization (use recruiter-lens).

Plan SEO topic clusters and eliminate keyword cannibalization so exactly one page wins each query. Use when mapping pillar and satellite pages, deciding which page should rank for a keyword, resolving internal competition between pages, writing title/H1/meta to avoid primary-keyword overlap, assigning cluster ownership from Search Console or from a sitemap when GSC access is missing, de-conflicting mixed-language pages with hreflang, or building keyword-research, on-page, and link-authority plans for organic growth. Not for crawl and index health (use seo-crawl), rewriting the copy once ownership is settled (use keyword-resolver), or mass page generation (use programmatic-seo).

Resolve keyword cannibalization with Keyword-Guard, then generate alternative titles/H1/meta descriptions with Wordsmith. Use when the user wants to eliminate internal competition between pages, assign query ownership from Search Console or sitemap, de-conflict mixed-language pages with hreflang, or build cluster maps — and wants both the diagnosis and concrete rewrite alternatives. Chains keyword-guard (audit) → wordsmith (alternatives) in one workflow. Not for technical SEO diagnosis (use seo-audit) or mass page generation (use programmatic-seo).

Drafts the policy pages a small product needs before it takes money or data (privacy policy, terms of service, refund policy, cookie notice) from a fact sheet about what the product actually collects, stores, shares and charges, then checks the drafts for placeholders, missing sections and facts nobody supplied. Use when someone needs a privacy policy, terms, refund or cookie page for a site, app, SaaS, newsletter or digital product, asks what legal pages they need before launch, or wants existing policy pages checked against what the product really does, even if they never say "legal". Not legal advice and not for contracts, NDAs or employment documents (use a lawyer), and not for the launch plan itself (use solo-income-blueprint) or the site build (use site-prompt).

Platform-native content packs: Twitter/X threads, LinkedIn posts, YouTube and TikTok scripts, Instagram captions, newsletter issues, blog articles, podcast outlines, 30-day calendars and repurposing chains, each with hook, body and call to action, built from a brief about your platform, niche, audience and goal. Use whenever someone asks for social or audience content: write me a thread about X, a LinkedIn post that brings leads, turn this into ten posts, plan next month of content, a YouTube script people actually finish, or they bring a blog, newsletter or podcast idea that needs an asset per platform, even if they never say content strategy. Not for copy that sells a product (use sales-copy), rewriting text the user already wrote (use wordsmith), SEO cluster and cannibalization planning (use keyword-guard), or career documents (use career-track).

Audit and fix real-world web app performance before launch using four independent read-only subagent audits covering images and fonts, frontend bundles and rendering, database queries and APIs, and server, network and user-flow measurement. Use when asked to find and fix slow pages, heavy bundles, N+1 queries, missing indexes, unoptimized images or fonts, or launch-readiness performance in an application the user owns or is authorized to maintain. Requires same-condition Lighthouse baselines and post-fix medians so claims are measured, not implied by code review. Speed only. Not for authorization and abuse testing (use pre-ship-security-red-team) or exposed secrets and source files (use web-preflight).

Authorized, evidence-driven security red-team pass over a web app you own before launch, run by three parallel read-only subagents with synthetic test accounts. Covers anonymous data access, IDOR cross-user object access, self-service role escalation, password-reset token reuse, UI-disabled request bypass, quota and free-tier bypass, concurrent quota races, hostile form input, open redirects, stored XSS, SSRF in URL-fetching features, private-response and cache isolation, preview deployment exposure, and prompt injection through ingested AI content. Then fixes each failure, reruns with a negative control and emits a before/after verdict table. Use when someone says "play the bad guy before we ship", "red team my site", "test my auth and rate limits", "is my vibe-coded site safe to launch", or after adding login, roles, credits, uploads, or AI ingestion. Local or isolated environments only. Not for production, or for a passive exposure check (use web-preflight).

Production-ready code delivered as a complete artifact: scripts, bug fixes with the root cause, code reviews, API integrations, SQL of any complexity, automation jobs, scrapers, data analysis in pandas or R, technical documentation and architecture write-ups, each with setup steps, error handling and a plain-English explanation of every section. Use whenever someone describes something they want code to do, pastes an error or a snippet to fix or review, asks for a query, a scraper, an automation or a system design, even if they never say write code or name a language. Not for auditing a live web app before launch (use pre-ship-performance-audit, pre-ship-security-red-team), or for the business decision behind the build (use strategy-memo).

Turn a set of user-supplied prompts into a finished, tested agent skill. Runs an intent interview and overlap check, drafts with draft-skill, writes test cases, tunes with tune-skill, and optionally names and lists the repo with findable. Use whenever the user has a list of prompts, instructions, or example requests and wants them packaged as a reusable skill, even if they only say "make a skill from these", "turn these into a skill", or paste a batch of prompts with no further instruction. Not for editing an existing skill (use tune-skill), creating a skill from a plain description with no prompts (use draft-skill), or finding existing skills for a task.

Release a skill built from prompts. Use after prompts-to-skill has drafted and tuned a skill, to package it as a zip, run its test cases and record real results, check install and discovery in a fresh session, write the repo README and version, and hand the user an install checklist. Use whenever the user has a finished skill folder in context and says "package this skill", "release it", "ship it", "make it installable", "run the tests on this skill", or "is it ready to install", even if they don't name a step. A bare "publish this to GitHub" with no skill to package is not this job. Not for drafting or tuning a skill (use prompts-to-skill or tune-skill), for repo discoverability copy (use findable), or for pushing existing clones (use upstream-check).

Sets up, audits and repairs a manifest-driven CI/CD pipeline that publishes a collection of agent skills to a Cloudflare Pages site, with a generated page per skill. Use when someone asks for CI/CD or automation for a skills site, wants a new skill repo to appear on the site with its own page, asks why a skill is missing or its page is stale, or needs a red publish run diagnosed and a bad deploy rolled back. It installs one GitHub Actions workflow with check and publish jobs, a review-PR or auto-sync publication gate, generated hub rows, discovery indexes, sitemap and pages, and gates on both sides of the deploy. Not for releasing one skill through the marketplace, tarball or plugin channels (use publish-release-gate), checking unpushed work or listing drift (use upstream-check), deciding whether a build is safe to launch (use release-gate-chair), or writing and tuning a skill (use draft-skill or tune-skill).

Review resumes, CVs, cover letters and LinkedIn profiles the way recruiters commonly screen them, and prepare candidates for interviews. Use when the user shares or asks about job-search materials: why a resume is not getting interviews, a quick first-impression scan, ATS compatibility or keyword fit against a job description, tailoring coverage against a role, gap analysis, weak bullet points, predicting interview questions, or auditing a LinkedIn profile. Also covers using AI honestly in a job search. Never invents achievements or metrics. Not for producing finished application material (a rewritten resume, new summaries, a full LinkedIn rewrite, interview answer scripts): use the applicant-kit skill, which takes this skill's findings as input. Not for running the search itself (finding roles, the application tracker, follow-ups, interview briefings): use the application-pilot skill. Not for prose editing (wordsmith, if installed) or producing the final .docx or .pdf (the file skills).

Runs the three pre-launch passes in order — web-preflight for external exposure, pre-ship-security-red-team for application behaviour in an isolated environment, pre-ship-performance-audit for measured speed — records each lane's own verdict and findings into one ledger, and computes a single release posture (HOLD or ready) with a script rather than judgement. Use when someone asks "is this safe to launch", "ready to ship", "should we deploy", "give me a go/no-go", or before any deploy of a site or client app you own or are authorized to test. Three verdicts stay separate; a pass in one lane never covers another. Variant: this fork runs the lanes itself and copies their finding tables verbatim; it is the default release gate. Not for a single-lane check (invoke that skill directly), for reports that already exist (use release-gate-validator), for penetration testing of production, or for performing the deploy.

Runs the three pre-launch passes in order and has each lane write its own rows into one release-gate ledger through a validating recorder, instead of a chair copying tables after the fact. Produces three separate verdicts and one release posture (HOLD or ready) computed by script. Use when the three lane skills carry the ledger patches below, or when someone asks for the ledger variant of the release gate, for a site or client app you own or are authorized to test; a plain go/no-go belongs to release-gate-chair. Ships the three upstream patches that teach web-preflight, pre-ship-security-red-team and pre-ship-performance-audit to append their own rows — unapplied, because those are published skills. A pass in one lane never covers another. Not for a single-lane check, not for the default go/no-go (use release-gate-chair), not for penetration testing of production, not for performing the deploy.

Turns three pre-launch reports that already exist into one release posture (HOLD or ready) without running any of them. Ingests the security red team's on-disk results table mechanically, takes the preflight and performance tables as verbatim pastes, and computes the posture with a script. Use when the lanes were run earlier or by someone else, or when a deploy decision is needed now from those reports and re-running three skills is not affordable. Runs no attacks, no probes, no measurements, no subagents. A lane's pass never covers another lane, and an absent lane is reported absent. Not for producing the reports in the first place (use release-gate-chair), and not for deploying.

Decision-grade research: topic overviews, competitor and industry analysis, literature reviews, fact checks with conflicting sources flagged, synthesis of expert positions, trend scans, due diligence on a partner or investment, interpretation of numbers found in public sources and one-page executive briefs, with every claim tied to a dated source read in the session and unknowns named instead of filled in. Use whenever someone asks what is known about a subject: research X for me, who else does this, is this claim true, what is the market size, brief me before I meet them, summarise this literature, even if they never say research. Not for turning findings into a recommendation and roadmap (use strategy-memo), for analysing the user's own dataset (use data-readout), or for being taught a topic step by step (use topic-primer).

Direct-response conversion assets: long-form sales pages, launch and abandoned-cart email sequences, Facebook, Google and LinkedIn ad variations, landing pages, product descriptions, video sales letter and webinar scripts, upsell flows and testimonial frameworks, each with headline, subheadline, body and call to action, built from the product, the buyer, their awareness level and the conversion goal. Use whenever someone needs copy that sells: write a sales page, an email sequence for my launch, ad copy for Facebook, my landing page does not convert, a VSL script, recover abandoned carts, or they paste a product brief expecting buyer-facing text, even if they never say copywriting. Not for organic audience content (use platform-content), editing text the user already wrote (use wordsmith), or pricing and go-to-market decisions (use strategy-memo).

Use when you need evidence about the technical SEO state of a site you own or are authorized to crawl, covering HTTP statuses and redirects, indexability (meta and header robots), robots.txt, canonical tags, titles, meta descriptions, H1s, internal links, sitemap coverage, and static JSON-LD. Runs a bounded, read-only crawl and returns deterministic findings JSON with stable IDs and explicit INCONCLUSIVE checks, so an SEO audit or content change starts from measured evidence instead of guesses. Trigger phrases include "crawl the site for SEO", "check sitemap coverage", "find broken internal links", "why is this page not indexed", "check titles and metas", "canonical audit", and "pre-audit crawl". Not for Search Console clicks, impressions, rankings or keyword volumes; Core Web Vitals, Lighthouse or bundle performance; release security and exposed files; content rewriting or keyword strategy.

Audit SEO health with seo-crawl (and seo-audit when installed), then generate optimized alternatives with Wordsmith. Use when the user wants a full technical + on-page SEO diagnosis followed by concrete rewrite alternatives for titles, meta descriptions, headings, content gaps, and internal linking — and wants both the audit findings and polished fixes. Chains seo-crawl (measured evidence) → seo-audit (optional diagnosis) → wordsmith (alternatives) in one workflow. Not for keyword cannibalization resolution (use keyword-resolver) or mass page generation (use programmatic-seo).

Writes one self-contained, paste-and-build prompt for a website from a topic. Narrows the topic to a concrete subject, derives the concept from that subject's world, builds and verifies the site in a scratch folder, then embeds every file verbatim with a pinned stack, final copy, code-native visuals and the checks that were actually run. Use when the user asks for a website prompt, one-shot build prompt, mega-prompt or SELENE-style prompt for a new site, landing page or one-pager. Not for building, hosting or reviewing the site itself, not for redesigning an existing page, and not for authoring skills.

Takes a whole collection of agent skill repos from scattered to published, consistent and self-maintaining, in gated phases: inventory every repo, reconcile site, marketplace and README listings, test and tune the family, close workflow gaps, wire publishing so new skill repos go live on their own, bundle a plugin with agents, add trigger tests and directory listings, and leave a maintenance loop and a map. Each phase hands off to the skill that owns it and ends in PRs. Use when someone's skills site and GitHub disagree, wants all their skill repos audited or cleaned up, wants new skills to publish themselves, or wants that cleanup repeated on another org, even if they never say estate. Not for one skill (use tune-skill), one listing (use findable), the site pipeline alone (use publish-pipeline), or weekly upkeep (use caffeinity-routing's Descale brew).

Launch a new skill end-to-end. Create it with draft-skill, validate and tune it with tune-skill, then audit its discoverability and polish it with findable-polish. Use when the user has an idea for a skill and wants to go from zero to published, with triggering, validation and discovery surfaces all handled in one pipeline. Chains draft-skill, tune-skill and findable-polish sequentially. Not for editing existing skills (use tune-and-polish), and not for a batch of prompts the user already wrote (use prompts-to-skill).

Decide whether an existing agent skill should become a Claude Code mod (code that runs inside Claude Code on events like tool calls, prompts, turns and compaction), and if it should, convert it into a validated mod plugin that keeps the original skill bundled. Use whenever the user wants to turn a skill into a mod, "mod-ify" a skill, make a skill run automatically before git push or deploys or on every prompt or after edits, give an existing skill a pane, status line, band or slash command, or asks "can this skill be a mod?", "should this be a mod or a skill?", even if they never say "hooks". Needs Claude Code 2.1.287 or later. Not for writing a brand-new mod with no source skill (use plugin-authoring), writing a new skill (use draft-skill), improving a skill's wording or triggering (use tune-skill), or plain skill packaging (use skill-creator).

Design a realistic one-person online income plan from the user's skills, time and budget: one recommended business model, one offer, a demand check, a content and sales path, automations, a minimal tool stack, and a 7-day launch plan. Use when someone asks how to make money online or with AI tools, what to sell, which business model fits them, how to turn a skill they already have into income, how to get a first paying customer, or wants a step-by-step plan for a side business or one-person business. Not for job hunting (application-pilot, applicant-kit), SEO for an existing site (seo-audit), or privacy and legal pages (paper-trail).

Business strategy written as a decision document: competitive analysis, market and customer research, business model canvas, growth and pricing strategy, go-to-market plan, SWOT ranked by actionability, ninety-day, one-year and three-year roadmaps, board presentations and pivot assessments, structured as situation analysis, recommendation and execution roadmap. Use whenever someone faces a business decision: should I enter this market, who are my competitors, how should I price this, a go-to-market plan, build me a roadmap, is this worth more investment or a pivot, I need a board deck, even if they never say strategy. Not for pressure-testing a decision you already made (use crash-test), the copy or pages that execute it (use sales-copy, platform-content), or the numbers behind it (use data-readout).

A personal system for acquiring a skill: decomposition into the minimum viable components, a beginner-to-practitioner roadmap, the mental models that unlock a subject, Feynman-level explanations, deliberate-practice routines, knowledge-gap audits that surface wrong assumptions, spaced-repetition schedules, application exercises and honest mastery assessments of where the learner actually is. Use whenever someone has a skill already in progress and needs the structure around it, asks how to practise, how long a skill takes, what they are missing, for a study plan or revision schedule, or says they keep forgetting what they read, even if they never say curriculum. Not for being taught the topic itself concept by concept (use topic-primer), for career positioning and moves (use career-track) or for course content written for a class of students (use course-builder).

Use when the user names a topic and wants to be taught it — from absolute zero, step by step, concept by concept, through a full beginner-to-expert curriculum, with a personal tutor that diagnoses and adapts, in how experts think and which mental models to use, in avoiding common beginner mistakes, tested by a strict examiner, or explained in ultra-simple terms (ELI5). Triggers: "teach me X", "learn X from zero", "curriculum/roadmap for X", "quiz/test me on X", "how do experts think about X", "mistakes when learning X", "explain like I'm 5", "simple explanation", or /topic-primer X. Not for file-format conversion (use wordsmith), resume screening (use recruiter-lens), or SEO audits (use seo-audit).

Tune a skill's trigger and boundaries with Tune-Skill and then polish its discoverability with Findable-Polish, in one pass. Use when the user wants both halves at once, meaning a triggering or sibling-overlap fix AND polished repo description, topics, README or site page; chaining them means the listing is never polished for a skill whose triggering is still broken. Not for creating skills from scratch (use draft-skill), and not for a triggering-only or listing-only request, which belong to tune-skill and findable-polish alone.

Audit and fine-tune an existing agent skill so it triggers when it should and hands off cleanly to its siblings. Reads the real SKILL.md and references, checks the description, sibling overlaps, stale names and counts, handoffs and command routing, makes minimal evidenced edits, validates them with a script, writes near-miss test cases, and re-checks the whole family against a saved baseline. Use whenever the user wants to fine tune, improve, tighten, debug or review a skill, says a skill is not triggering or fires on the wrong requests, wants two skills to work together or split cleanly, asks whether anything in a set of skills got worse, or pastes a SKILL.md or repo link and asks what is wrong with it, even if they never say "audit". Not for creating a skill from scratch (use draft-skill) or from a list of prompts (use prompts-to-skill), or for making a skill run automatically at a moment such as before git push (use skill-to-mod).

Find local work that never reached its remote and push only what is safe, then check whether published listings agree with the repos. Fetches a folder of git clones, classifies every repo as push-ready, level, behind, diverged, no-upstream or no-remote, screens the outgoing diff for credential-shaped lines, and with --push fast-forwards only the clean ones and never forces, uses --no-verify, or touches a diverged repo. A second checker counts a catalogue wherever it is published (site index, homepage, marketplace, READMEs) against the repos that ship a SKILL.md, and names what drifted. Use when the user asks what has not been pushed, whether clones are behind upstream, to push a batch, or why a site and a README list different numbers than GitHub. Not for writing or tuning a skill (use draft-skill or tune-skill), making a repo discoverable (use findable), or deciding whether a build is safe to deploy (use release-gate-chair).

Audit upstream drift and listing mismatches with upstream-check, then generate polished fixes with Wordsmith. Use when the user wants to reconcile repo state with published listings (marketplace, site index, README), fix descriptions or topics that drifted, or clean up skill boundaries after a push — and wants both the diagnosis and concrete rewrites. Chains upstream-check (audit) → wordsmith (alternatives) in one workflow. Not for creating skills from scratch (use draft-skill) or tuning trigger language only (use tune-skill).

Use when shipping, deploying, publishing, launching, reviewing, or pre-flighting a website or API and you want a lightweight check for accidental exposure of secrets, source or configuration files, admin and debug surfaces, unsafe redirects, missing security controls, and misleading 200 responses. Trigger phrases include "before we ship", "security check", "preflight", "audit the site", "check the production URL", "is my .env exposed", and "why does /.env return 200", including when access-log or analytics evidence is supplied. Checks only sites the user owns or is authorized to assess, with bounded GET and HEAD requests and a fixed compact path set; distinguishes genuine resources from SPA catch-all and blanket WAF responses. Not for penetration testing, exploitation, credential testing, broad port scanning, or application-code review; use a dedicated authorized security-testing or code-review workflow instead.

Write, draft, rewrite, edit, proofread, polish and improve text while preserving the author's intended meaning, facts and voice. Use when the user wants clearer, shorter or better-toned writing; grammar, structure or flow fixes; emails, website or marketing copy, headlines, outlines or summaries; a different reading level; an accessibility, inclusivity or accuracy review; or content prepared for publication, even if they never say "edit". Also handles /wordsmith commands such as draft, critique, audit, tighten, tone and final. Not for resumes, CVs or LinkedIn profiles (use recruiter-lens if installed), not for code or code comments, and not for building .docx or .pdf files (write the text here, then use the file skill).