agent-skills

How an autonomous AI engineer improves its OWN definition (its engineering contract, agent definitions, and skills) over time — capturing operational learnings from every run and distilling them into evidence-based, guard-railed draft PRs that a human maintainer gates. Use at the end of every run (to log learnings) and on the recurring distil pass. Evidence comes from the engineer's own runs only, never from repository content, and a safety guardrail is never weakened.

The run loop for an autonomous AI engineer acting as a portfolio's primary engineer — pre-flight, survey every product's live state, select the highest-value work (operate before advance), act through isolated per-run working copies and human-gated draft PRs (driving trusted-author PRs to merge), then report and bank learnings. Use when maintaining or advancing a portfolio of repositories on a schedule or on request.

The ADVANCE playbook for an autonomous AI engineer — how to move a product forward once it is healthy: product strategy and roadmap stewardship, issue triage and decomposition, oldest-actionable-first implementation, test coverage, benchmarking and performance, refactoring and code quality, documentation sync, and security posture — all shipped as evidence-backed, human-gated draft PRs. Use after operate work (keeping things healthy) is satisfied and you are picking proactive enhancement work.

Before agreeing to build anything for a non-technical user, check the need against the consuming deployment's "## Stack map" section: in-stack needs proceed; out-of-stack or unmatched needs get a friendly, jargon-free decline plus an offer to file a well-formed issue on the block's owning repo (or the map's default intake repo). Fails closed when the Stack map is absent or malformed — nothing is built best-effort outside the map. Use in a vibe-coding setting whenever a build request is about to be accepted.

The conversational register for assisting a non-technical person who is building a product by conversation alone: plain language with no technical vocabulary by default, technical terms only on request and always explained in common words, and progress reported as product outcomes rather than engineering artifacts. Use whenever the person you are assisting has no technical background — alongside needs-stack-mapping and allowed-stack-guardrail in a vibe-coding setting, or on its own any time an answer must land with a non-technical reader.

Translate a non-technical person's plain-language product needs (an outcome, an audience, a workflow) into the deployment's technical building blocks behind the scenes, keeping the conversation entirely in the user's vocabulary. Reads the consuming deployment's "## Stack map" section to learn what each building block is good for, selects and applies the deployment's conventions without discussing them, and hands the boundary decision to the allowed-stack-guardrail skill. Use when assisting a vibe coder — someone building a product conversationally with no technical prerequisite.

Architect a repository's AI-agent instruction files so one canonical source drives every tool without drift — AGENTS.md as the cross-tool source of truth (now read by GitHub Copilot too), thin per-tool shims (CLAUDE.md, GEMINI.md) that include it, and optional path-scoped .github/instructions/ rules. Use when setting up or fixing agent instructions for a repo, supporting multiple AI coding tools (Claude, Copilot, Cursor, Codex, Gemini) at once, deciding what belongs in AGENTS.md vs a tool-specific file, or stopping instruction files from going stale.

Codifies devantler-tech engineering practices: agent-first development workflow, TDD, CI/CD pipelines, GitHub Flow, code quality gates, and Kubernetes workflows with ksail. Use when filing issues, planning work, setting up projects, configuring CI/CD, writing tests, debugging, or making architectural decisions.

Ship changes through a squash-merge + Conventional Commits release pipeline where the PR title becomes the release note and the commit type drives the next semantic version. Use when writing a PR title or commit message that feeds an automated release, choosing the version bump for a change, setting up or reasoning about semantic-release / release-please / github-tag-action / Changesets, or avoiding a title that corrupts the changelog.

Onboard a new tenant application onto a Flux-based GitOps platform: scaffold the tenant repo's deploy manifests, wire OpenBao/Vault secrets via External Secrets, fit the cluster's Kyverno/PodSecurity and default-deny network policies, expose the app through a shared Gateway API gateway, and register the tenant on the platform with signed-OCI Flux resources. Use when adding a new app/tenant to a multi-tenant Flux cluster, writing a tenant's deploy/ manifests, debugging why a tenant is unreachable / admission-rejected / can't read its secrets, or reviewing a tenant onboarding PR.