gc-ai-tools

Imported from hgsantana/ai-tools (skills/gc-ai-tools/SKILL.md). Install upstream with npx skills add hgsantana/ai-tools --skill gc-ai-tools. Copyright stays with the author.

<session_workflow> Parse requested Google Cloud inspection, billing query, or resource operation. Derive a kebab-case {TOPIC}.

<step id="2" name="read_exploration">
  Run read-only inspection queries freely:
  - Project context: `gcloud config list`, `gcloud projects list`, `gcloud projects describe {PROJECT}`.
  - Services and assets: `gcloud {SERVICE} list`, `gcloud {SERVICE} describe`, `gcloud asset search-all-resources`.
  - Billing and monitoring: `gcloud billing accounts list`, `gcloud logging read`, `gcloud monitoring`.
  Prefer `--format="table(...)"` or JSON piped through `jq` for concise output.
  Send bulk log or fact collection to `<template role="mechanical-discovery">` from `<dispatch_templates>`, substituting {COMMANDS} and {TOPIC}.
</step>

<step id="3" name="mutation_guardrail">
  Identify if the command creates billable assets, mutates configurations, or deletes cloud resources.
  Present each mutation as an explicit approval request to the user per USER-AGENTS `<security_guardrails>`:
  - State exact command, target project/resource, reason, and cost or blast-radius impact.
  - Execute only after explicit affirmative user approval.
</step>

<step id="4" name="report">
  Write detailed inventories, cost breakdowns, and logs to `dev/tmp/{TOPIC}.md`.
  In chat (user's language), provide the direct answer, the report path, and any pending approval request.
</step>

</session_workflow>

<dispatch_templates> Default worker: run read-only gcloud commands and collect output. {COMMANDS} {TOPIC} Execute the read-only gcloud commands listed in {COMMANDS}. Save formatted command outputs to dev/tmp/{TOPIC}.md. Return command list, exit codes, and output path. Read-only queries only. Never execute mutating commands. </dispatch_templates>