ai-pgi-dev-kit

Claude Code skills for PGI development and operations.

Skills

Infrastructure

Skill	Description
infra-portforward	Port-forward to K8s services (RabbitMQ, PostgreSQL, APIs)
infra-pods	Pod status, diagnostics, events, resource usage
infra-logs	Pod log analysis and error searching
infra-deploy	Production deployments
infra-migrations	Database migrations in remote environments

RabbitMQ

Skill	Description
rabbitmq-list-queues	Queue status, DLQ counts, consumer health
rabbitmq-triage-dlq	Diagnose why messages fail in DLQs
rabbitmq-reprocess-dlq	Reprocess DLQ messages via shovel

Database

Skill	Description
db-query	Read-only SQL queries from within pods

Jira

Skill	Description
jira-create-bug	Structured bug reports with root cause analysis
jira-clarify-ticket	Detect ambiguities in ticket specs

Analysis

Skill	Description
ticket-analysis	Comprehensive ticket analysis (Jira + Figma + docs + code)
analyze-design	Extract data fields and states from Figma designs
analyze-docs	Search Confluence/Notion for documentation
analyze-data-model	Trace entity fields through the codebase

Meta

Skill	Description
creating-skills	Guide for authoring new skills

Docs

Document	Description
Ticket Analysis Workflow	Step-by-step workflow for analyzing tickets before implementation (Jira + Figma + Confluence/Notion + codebase)

Installation

Clone the repo and symlink the skills into your Claude Code personal directory:

git clone <repo-url> && cd ai-pgi-dev-kit

for dir in skills/*/; do
  name=$(basename "$dir")
  ln -sfn "$(pwd)/$dir" ~/.claude/skills/"$name"
done

The symlinks make ~/.claude/skills/ point directly to the repo files. This means:

Action	What happens
Use a skill	Claude Code reads ~/.claude/skills/X which resolves to the repo via symlink
Edit a skill	You edit the file in the repo (or via the symlink — same thing)
Share changes	git commit && git push from the repo
Receive updates	git pull — symlinks keep pointing to the same files

No manual copying needed in either direction.

Contributing

Adding a new skill

1. Create a folder under skills/ with a kebab-case name.
2. Add a SKILL.md file with the required frontmatter (name, description) and instructions. See the creating-skills skill for authoring guidelines.
3. Optionally add references/, scripts/, or assets/ subfolders for supporting files.
4. Update the skills table in this README.
5. Submit a PR with a clear description of the skill's purpose and trigger conditions.

Improving an existing skill

1. Test the current behavior by invoking the skill in Claude Code.
2. Make your changes to SKILL.md or supporting files.
3. Verify the skill still triggers correctly and produces the expected output.
4. Submit a PR explaining what changed and why.

General guidelines

• Keep skills focused on a single workflow — avoid catch-all skills.
• Follow the progressive disclosure principle: load only what's needed, when it's needed.
• Use references/ for detailed docs that the skill loads on-demand, rather than embedding everything in SKILL.md.
• Write trigger descriptions that are specific enough to avoid false activations.
• Include both English and Spanish trigger phrases in the description field where applicable.