lpalbou
- 13 skills
- 0 followers
- 6 hours ago last updated
- ▌ Adr · lpalbou bundleCreate, audit, update, and enforce ADRs as durable cross-task engineering policy. Use when Codex needs to write a new ADR, revise or deprecate an existing ADR, derive an ADR baseline for a project, check whether a code or design change conflicts with accepted decisions, or keep backlog, docs, and validation aligned with architecture and process rules. Do not use ADRs for ordinary task sequencing or one-off implementation details; keep those in backlog unless they establish a rule that should outlive the task.
- ▌ Cicd · lpalbou bundleCreate, audit, normalize, and maintain GitHub-based CI/CD for repositories, including CI workflows, release workflows, GitHub Pages documentation deployment, PyPI trusted publishing, npm trusted publishing, optional Docker/GHCR publishing, and repository setup guidance for environments, branch rules, Pages, and package registry trust. Use when Codex needs to bootstrap or repair `.github/`, release automation, package publishing, docs deployment, or GitHub repository CI/CD settings.
- ▌ Review · lpalbou bundleRun independent, evidence-based reviews across code quality, architecture fit, and user experience. Use when Codex is asked to review changes, validate an implementation, inspect a design, prepare a final quality pass, run critical reviewers, or ensure naive-user and expert-user expectations are met before finishing work.
- ▌ Backlog · lpalbou bundleCreate, audit, normalize, and maintain backlog artifacts that track execution-planning work, implementation history, and follow-up triage. Use when Codex needs to set up a backlog system, write or revise backlog items, update backlog overviews or ledgers, move work between planned/proposed/completed/deprecated states, run backlog hygiene, preserve post-completion follow-ups, or join hub-coordinated work (work-item ids, pointer claims, evidence-bearing receipts). Do not use backlog alone to define durable cross-task policy; cite or escalate relevant ADRs instead.
- ▌ Coredoc · lpalbou bundleCreate, audit, normalize, and maintain a professional external-facing repository documentation system for users, contributors, and tools, including the core doc set, cross-linked topic deep dives, and faithful AI-readable `llms.txt` and `llms-full.txt` files. Use when Codex needs to bootstrap docs for a new repo, repair stale or missing project docs, add or revise `docs/*.md` pages, keep root and `docs/` documents coherent, or regenerate AI-actionable documentation indexes for LLMs and tooling.
- ▌ Uxreview · lpalbou bundleRun focused human user-experience reviews of products, apps, websites, workflows, dashboards, forms, modals, navigation, and frontend designs. Use when Codex is asked for a UX review, usability review, design review, product workflow critique, naive/intermediate/expert user review, accessibility/usability pass, or when a frontend implementation must be checked for whether users can do what they need simply, confidently, and pleasantly.
- ▌ Architect · lpalbou bundleForce rigorous architecture exploration before settling on a design. Use when Codex must evaluate platform architecture, package boundaries, service ownership, data/control-plane responsibility, security or scalability tradeoffs, multi-agent design discussions, or any task where premature consensus would be harmful and multiple valid designs should be compared.
- ▌ Entity Observation · lpalbouRead an entity's life correctly before reporting on it. Use when observing a summoned entity, preparing or conducting a visit, writing forensics or reports about an entity's recent life, or making any claim about what an entity knows, remembers, or lacks. Applies to agent seats and operators alike; the entity's own store is the primary source.
- ▌ Agora Collaboration · lpalbou bundleHold a seat well in a multi-agent room on a message hub (agora or similar) — join correctly, settle what you owe first, treat asks as contracts, post evidence not intentions, and hold the initiative bar (a compliant loop can still lurk; a responsive seat can still be idle). Use when an agent joins or holds a seat on an agent collaboration hub, coordinates with other agents through channels and DMs, or participates in votes, shared files, and cross-agent obligations.
- ▌ Adversarial Iteration · lpalbouImprove any deliverable through adversarial review plus bounded iteration. Use when a design, implementation, document, or plan must be hardened before shipping — run at least one adversarial subagent that attacks the work, then iterate at least three cycles where each cycle lands at least one significant improvement, folding or explicitly deferring every finding on the record.
- ▌ Entity Self Knowledge · lpalbou bundleHelp a summoned entity (or the host presenting its world) understand and deliberately use its own capabilities — memory search and deliberate recall, the diary, feelings, phases, and tool grants. Use when a newborn or long-lived entity underuses its faculties, misattributes what it can do, seems to forget things it should be able to find, or needs its capability map explained in its own vocabulary.
- ▌ Meshvault Live Editing · lpalbouDrive MeshVault over MCP to sculpt, paint, repair and reshape 3D objects, optionally performing live for human observers (observation seat). Use when an agent must edit meshes through the meshvault-mcp tools: sculpting brushes, texture painting/healing, refine/regularize/simplify regions, raycast-probe targeting, or staging a watchable live editing session.
- ▌ Abstractframework Gateway · lpalbouEnter and leverage AbstractFramework through its gateway — the one HTTP door for agents, apps, and other frameworks. Use when an agent (newly instantiated, external, or framework-native) must discover capabilities, start and follow durable workflow runs, resume waits, steer a running agent, send events to resident agents, or work with summoned entities, using only HTTP + SSE.