Pedro Rosalba
Home

Claude Code skills

Tools I build for how I work

The skills I use on my own repositories, packaged as one Claude Code plugin. Rosalbito is the entry point; companion skills join in when a task calls for them.

Install

/plugin marketplace add PedroRosalba/rosalbito/plugin install rosalbito@rosalbito

Restart Claude Code inside the target repository.

Then:

/rosalbito add an integration test that runs the CLI on the sample CSV

If another skill is called rosalbito, use /rosalbito:rosalbito — it always works.

Skill 01 / 02 · Entry point

Rosalbito

One command for every engineering task. It classifies work by risk, runs an autonomous understand → plan → implement → verify → review loop, and delivers a verified pull request — or a precise question when a human is genuinely needed.

How it works

  • Overhead scales with the task: a typo gets near-zero ceremony; a critical change gets fresh-context adversarial reviewers and an engineering report.
  • Deterministic verification is the backbone — every check runs through a script that records exit codes as evidence; 'done' cites evidence.
  • Hard gates are mechanisms, not prompts: a PreToolUse hook denies force-pushes, pushes to main, production deploys, destructive SQL and PR merges.
  • The pull request is the human gate. Rosalbito never deploys and never merges.
  • State survives: a fresh session resumes an interrupted run from its state file.

Try it

/rosalbito add an integration test that runs the CLI on the sample CSV/rosalbito status/rosalbito  # resumes an interrupted run
GitHub

Skill 02 / 02 · Companion skill

Telemetry

A model-invoked companion skill for telemetry engineering: adding, changing and verifying traces, metrics and structured logs, and keeping a repo's telemetry context so the next change starts from what is already instrumented.

How it works

  • Ships in the same plugin as Rosalbito — no separate install.
  • No command to remember: the model reaches for it on telemetry-shaped tasks, such as “add tracing” or “add a metric for…”.
GitHub

Real run

risk=MEDIUM floor=MEDIUM human=visibility✓ plugin-validate — exit 0 — evidence #10✓ plugin-hook — exit 0 — evidence #12✓ plugin-hook-e2e — exit 0 — evidence #7✓ suite — exit 0 — evidence #11
Condensed from the evidence log of the run that packaged this plugin: each line is a check that ran, with its exit code. A run is only done when it can cite lines like these.