Audit documentation quality with weighted scoring across 12 dimensions (Diataxis coverage, completeness, freshness, readability, examples, accessibility, CI testing). Detects doc anti-patterns (Wall of Text, The Lie, Jargon Soup). Use when reviewing doc quality, auditing documentation, checking for stale docs, or saying 'audit docs', 'doc quality', 'documentation review'. Invoke with an optional file or directory path.
Generate and audit comprehensive project documentation from a Git repository. Produces standard documentation (CONTRIBUTING.md, architecture, troubleshooting) in Diataxis structure with zero AI markers by default. Use this skill whenever someone asks to document a project, generate docs, create a README, write API docs, or improve existing documentation. Also trigger when someone mentions "doc", "documentation", "readme", "explain this codebase", "onboard developers", or "make this project understandable".
Ghostwrite, rewrite, and audit everyday professional exchanges in the user's voice: emails, replies, chat messages (Slack/Teams/Discord), LinkedIn DMs, client and recruiter answers, follow-ups, thank-yous, refusals, bad-news notes, meeting requests, and short announcements. Enforces zero AI tells (no em-dash reflex, no AI vocabulary, no assistant structure), front-loaded asks, one message = one ask, target-language proofreading with correct diacritics, and calibrated light humor for follow-ups. USE BY DEFAULT — even without a trigger word — whenever creating, editing, rewriting, translating, or auditing ANY text the user will send or publish in their name: message, mail, réponse, annonce, petite annonce, post, commentaire, DM, description produit, texte de formulaire. If the task involves producing or modifying user-facing prose, load this skill first. Also triggers on 'écris un mail', 'réponds à', 'relance', 'relance-le', 'follow up', 'reformule ce message', 'humanize this', 'sounds like AI', 'trop formel',
Generate a production-parity resilience blueprint: test battery, troubleshooting runbook, agent-ready operations pack, failure-injection plan, and incident blackbox templates. Uses biological and physical reasoning — genome/contracts, membranes/boundary conditions, homeostasis/health checks, immune system/negative tests, stress-strain/failure budgets, phase transitions/resource cliffs, hysteresis/reruns, and memory cells/post-incident capitalisation. Use whenever the user asks to prevent prod bugs, make dev/staging closer to prod, design pre-prod checks, write a runbook, create a smoke test ladder, harden deployments, define N1/N2/N3 operations, govern agent autonomy, or turn incidents into durable anti-regressions.
Audit the ordering and layering of install/config/build steps using the Tower of Hanoi model: stable-and-expensive at the bottom, volatile-and-cheap on top, no step before its prerequisite, minimal redone work per change, and no responsibility displaced to a higher layer (what the code does not do resurfaces in Docker, kube.yml, or Ansible — often paid three times). Applies to Dockerfiles/Containerfiles (layer cache, image size), Ansible playbooks/roles (task order, handler restarts, idempotence), Kubernetes manifests (apply order, CRD/RBAC/ConfigMap sequencing), Rust workspaces and build systems (crate layering, incremental compile blast radius), CI job ordering, long documents (reader-cache ordering), and cross-layer stacks (code -> image -> orchestrator -> config management). Use when the user says 'hanoi', 'Tours de Hanoï', 'ordre d'installation', 'ordre des tâches', 'layer order', 'layer caching', 'Dockerfile cache', 'ça rebuild tout', 'build lent', 'image trop grosse', 'restart storm', 'apply order', 's
Audit shell scripts against Google Shell Style Guide + ops best practices. Scores 12 dimensions: strict mode coherence, error surfaces, logging, stderr hygiene, variable discipline, quoting, control flow, naming, CLI ergonomics, idempotency, namespace, and security. Goes beyond shellcheck — detects semantic anti-patterns invisible to linters (dead fallbacks under set -e, custom loggers vs logger(1), redundant package checks, env var injection in heredocs, missing getopts). Use when reviewing shell scripts, auditing bash code, checking deployment scripts, or saying 'audit shell', 'bash review', 'script quality', 'shell style', 'shellcheck not enough', 'review my script'. Also triggers on 'set -euo pipefail', 'getopts', 'shell injection', 'logger', 'bash best practices', 'google shell style'.
Audit and fix GitHub repository health: rulesets vs CI alignment, branch hygiene, PR lifecycle, release automation flow, permission issues, and transient CI failures. Detects misconfigurations that cause PRs to hang, CI to fail silently, branches to accumulate, and releases to stall. Use when the user says 'PR stuck', 'CI pending forever', 'branch cleanup', 'ruleset', 'release blocked', 'merge conflicts on sync-main', 'stale PRs', 'orphan branches', 'GitHub health', 'repo hygiene', 'required checks', 'path pruning', 'release-plz stuck', 'auto-merge not working', 'workflow scope', 'token permission'. Also triggers on 'gh api', 'rulesets', 'branch protection', 'status checks'.
Agentic migration architect for transforming Ansible, bare-metal, VM, shell-scripted, or systemd-root deployed middleware into an operable OCI rootless product using any OCI-compliant runtime (Podman, Docker rootless, nerdctl, or equivalent), systemd --user, declarative units (Quadlet, compose), explicit host bootstrap, a single operator CLI, executable gates, monitoring, and proven recovery. Uses a stratigraphic/metallurgical reasoning model: extract the bedrock contract, separate ore from gangue, map fault lines, forge target alloys, stress-test fracture surfaces, and remediate legacy tailings.