tedeuxx/tadeumendonca-skills

Run a slice through this loop's intentional design — its intake chain, state machine, iteration axis and inner-loop steps. Use when picking up a slice, proposing a change to the loop itself, or naming Agent Harness Engineering in public writing. Not the portable judgment (see engineering-standards), what "done" means (see definition-of-done), the CI/CD gates (see quality-gates), the permission zones (see scm), or what makes an item ready (see definition-of-ready).

Implement a BFF-on-Lambda backend as a reference pattern — the Hono modular monolith, cross-cutting middleware (errors, logging, metrics, tracing, audit, action types), Redis cache-aside, config/secrets, the generated OpenAPI contract + Postman tests, notifications, and OG-image + bot-rendering. Use when a project actually needs a BFF and its cross-cutting shape. Not for the AWS resources it runs on (see cloud-infrastructure) or the SPA calling it (see frontend).

Wire continuous integration for a <project> repo — workflows, required checks, runners, pinned actions, short-lived cloud credentials, the secrets standard, version and release workflows, AI review. Use when writing or changing a workflow, making a check required, granting a pipeline a cloud role, or a required check never reports. Not for branching or release rules (see scm), gate thresholds or the analysis tool (see quality-gates), or infrastructure state (see provisioning).

Provision AWS infrastructure in Terraform end to end, one section per service — network, identity and encryption, the config/secrets bus, data stores and cache, storage, compute, API and CDN edge, certificates, DNS, email and event fan-out, and observability. Use when provisioning any AWS resource, choosing between two services, or tracing a cross-service reference. Not for the CI/CD pipeline that runs plan and apply, branching, or the permission model (see devops).

Run the author's own completeness pass before opening a merge request — every requirement marked met or unmet, every new assertion mutation-checked, what the change made false, alternative spellings of anything parsed, and the gates run with real output. Use when a slice is finished and not yet submitted, or when anticipating both of the gate's lenses. Not for the definition of done itself (see definition-of-done), or the CI/CD gates it anticipates (see quality-gates).

Move a finished piece through its whole lane — the owner's selection, the drafting pair and its two-round bound, the reviewer that REPAIRS a copy defect in place, the held preview, release, the social pair — and say what the owner reads before a content merge: the preview mode (held preview or pending revision), never a Claude Artifact. Use when taking a piece from a ready Issue to live, correcting a published piece, or asking which step runs unattended and which waits on the owner.

Design or evaluate a Definition of Done, and apply THIS loop's own — the criteria a slice calls finished, plus the table naming which of them a gate proves and which nothing does. Use when building a DoD from scratch, checking a slice against this loop's criteria, or diagnosing why review still reads as subjective. Not for the CI/CD gates and thresholds themselves (see quality-gates), or a work item's readiness to be built (see definition-of-ready).

Set the bar a work item must clear before a builder can pick it up, generically and for THIS loop — what ready asserts and which of its criteria anything actually checks. Use when writing acceptance criteria, running intake, sizing a backlog, or diagnosing why in-progress work stalls on undecided scope. Not for what "done" means at delivery (see definition-of-done), or the state machine of who acts at each transition (see agents-configuration).

Operate DevOps for a <project> repo — GitHub Actions, Terraform Cloud, branching, the pipeline-only IaC floor, numeric SemVer, the Claude Code GitHub App, and SonarCloud (setup, CI step, gate wiring). Use when wiring a pipeline, granting CI a role, cutting a release, bumping SemVer, or a Sonar gate is red. Not for Terraform config (see cloud-infrastructure), state machine (see agents-configuration), the gate list Sonar sits inside (see quality-gates), or the pre-merge pass (see code-review).

Write or review documentation for any <project> repo — general docs (README, architecture pages, diagram choice, Markdown + Mermaid only) AND Architecture Decision Records (MADR format, the significance gate, the methodology/product library split, numbering and status, and the current-codebase rule for a reversed decision). Use when adding a README or architecture page, choosing a diagram type, reviewing drifted docs, writing or superseding an ADR, or judging architectural significance.

The owner's engineering preferences, portable to any project — two tiers (a non-negotiable floor, a risk-calibrated judgment layer), the twelve principles behind every call, delivered versus hygiene, and the human residual. Use when deciding how much rigor a change deserves, whether something counts as delivery, or what to do while blocked. Not this loop's machinery, state machine or WIP rule (see agents-configuration), and not what done means concretely (see definition-of-done).

Build the React + Vite SPA end to end — bootstrap and providers, routing, state ownership (React Query / Zustand / the Cognito SDK), the typed BFF client, auth and cosmetic UI gating, forms, cursor pagination, the custom Tailwind design system and Storybook, async UX states, markdown rendering, SEO, GA4 analytics, and Playwright E2E. Use when building or reviewing any SPA feature. Not for the AWS resources it calls (see cloud-infrastructure) or the API it calls (see backend).

Apply three main-session rules — the review a merge request gets for its type, what a loop mode may never change, how to ask the owner, and what the owner reads before a content merge (preview mode, never a Claude Artifact). Use when routing or asking. Use before dispatching a review, before any pool query or rite, and before any question, decision or action you put to the owner. Not for the state machine (see agents-configuration) or escalation's argument (see engineering-standards).

Run a planning poker estimation round, or diagnose why one produced theater instead of signal. Use when the agents estimate an item's sp:N weight, each dispatched in isolation with the median recorded, when sizing backlog items with a team, choosing between group-consensus estimation and a single person's gut call, or explaining why a session's numbers don't influence anything downstream. The agents of this method use it to estimate, per its Estimation section (see agents-configuration).

Provision infrastructure for a <project> repo through the pipeline only — remote state and runs, one workspace per environment, plan on the pull request and apply on merge, never from a laptop, infra-first ordering. Use when adding or changing infrastructure, wiring a plan or apply job, setting up remote state, or tearing something down. Not for infrastructure code patterns (the site project's cloud-infrastructure skill), workflow wiring and pipeline roles (see ci), or branching (see scm).

Judge or draft anything published in the owner's voice — articles, site copy, and LinkedIn/X teasers — against the calibration he ratified: the goal, the filter, the subject bound, the corpus evidence, the ranked title criteria and the teaser arc. Use when a piece is being drafted, or when a draft is being reviewed before it reaches him.

Apply THIS loop's CI/CD gate policy — the gate table per loop model, merge-class rules, thresholds (lint/typecheck zero, coverage ≥85%, contract/E2E, dependency and secret scanning, SAST) and the static-analysis gate (setup, mode, CI step, new-code thresholds). Use when calibrating gates, deciding who merges a class, or diagnosing a red quality gate. Not for what makes a slice complete (see definition-of-done), the pre-merge pass (see code-review), or workflow wiring (see ci).

Run source control for a <project> repo — PRs, merge commits only, PR-only path to main, rc/next integration, labels, numeric SemVer releases, repo settings, the forge-act permission floor, the AI assistant app. Use when branching, opening or merging a PR, labelling, releasing, checking repo settings, or deciding if a forge act may run. Not for workflow wiring (see ci), the quality gate (see quality-gates), infrastructure (see provisioning), or the state machine (see agents-configuration).

Apply the platform's working-files and shell discipline in any <project> repo — where scratch files go, one atomic Bash call per invocation, the gh --repo flag position, and --body-file for any multi-line PR/issue body. Use whenever a persona writes a working file, runs a shell command, or composes a PR/issue comment. Not for the loop's state machine (see agents-configuration) or what a merge verdict must contain (see quality-gates).