README.md

Skills

Reusable skills installed via agent-config-kit (agent-kit) — see the repo-level Quick Start for installing agent-kit and applying agent-config.toml.

Skills are organized by category. Each skill lives in skills/<category>/<skill-name>/SKILL.md and carries YAML frontmatter with name, description, category, and tags fields.

Categories

Category	Description
python/	Python tooling and dependency management
dagster/	Dagster pipeline development with dg
infrastructure/	Pulumi IaC and Vault secrets
containers/	Docker image builds
workflow/	Cross-cutting process conventions
process/	GitHub and external service interactions

All Skills

Category	Skill	Description
python	uv-python-workflow	Use uv exclusively for all Python env & dependency management
python	cyclopts-cli-scripts	Use cyclopts for CLI scripts; place in bin/
python	drf-api-performance	Fast DRF APIs: nesting limits, pagination count query, prefetch choice, no ORM in serializers
dagster	dagster-code-location-structure	dg_projects/ layout, asset/sensor placement, one-at-a-time migration
infrastructure	pulumi-modify-existing	Modify existing stack entrypoint; never create new files; preserve assumeRole
infrastructure	vault-k8s-auth	Wire Vault K8s auth via hvac; never hardcode role or mount path
containers	docker-uv-image-builds	mitodl/<service> naming, git short-ref tags, relocatable uv venvs
workflow	validate-before-commit	Run pre-commit → mypy → pulumi preview proactively before declaring done
workflow	creating-skills	Create a new skill: frontmatter, category placement, progressive disclosure, index updates
workflow	extract-style-profile	Profile how a person, team, or repo codes and communicates from git/GitHub history; outputs CLAUDE.md/AGENTS.md rules with evidence
process	create-ol-github-issue	Create mitodl GitHub issues using org standard templates
process	create-ol-pull-request	Create mitodl pull requests using the org's standard PR template, after an independent goal/security review and a claim audit
process	create-ol-rfc-discussion	Write and post a structured RFC as a GitHub Discussion in mitodl/hq under the RFC category
process	generate-standup	Generate and post a daily standup from GitHub activity to the mitodl/hq Check-ins discussion
process	screenshot-pr	Capture UI changes for a PR with shot-scraper at desktop, tablet, and mobile viewports
process	dependency-updates	Triage and apply Renovate dependency updates safely across Python, JS/TS, Helm, Apt, and database ecosystems
process	dependency-pruning	Audit dependencies to find unused ones to remove and underused ones to vendor or rewrite
process	github-issue-triage	Audit open GitHub issues to identify stale, completed, or superseded items by cross-referencing the codebase (parallel subagents where available, sequential batches otherwise)
process	github-pr-triage	Categorize open PRs across an org by required action (needs first-pass review, has feedback, approved & ready to merge) and optionally act on them
process	address-pr-feedback	Fetch, categorize, address, and resolve GitHub PR review feedback, with pagination for large/long-running PRs
process	code-review	Review a diff, branch, path, or PR for correctness bugs, gaps against stated goals, security issues, and reuse/simplification/efficiency cleanups, with a verify-before-reporting pass
process	renovate-security-triage	Rank open Renovate PRs in your active repos by security urgency using advisory severity, CVSS, and EPSS (read-only)
process	deploy-verification	Verify a merged config/infra change actually took effect — CD pipeline, pod rollout, running config, before/after metrics, and unintended-environment scope
process	run-canary-locally	Get a passing local run of the ol-infrastructure Playwright canaries, without risking the account lockout, and read a failed run's trace
process	add-canary-journey	Add a canary journey to an existing property, or onboard a new property into the canary fleet (two list edits)

Platform capabilities and optional Pi extensions

agent-kit apply installs every skill into every detected platform, so a skill can't assume Claude Code's own tools. Where a skill benefits from delegation or a structured question, it says what to do when the capability is present and what to do without it (e.g. github-issue-triage runs its batches in parallel subagents when available, otherwise sequentially with the same evidence and report).

Stock Pi has no subagent tool, no AskUserQuestion-style questionnaire tool and no built-in /review command. These third-party Pi extensions add the first two and are picked up automatically by skills that check for the capability; they are optional, agent-kit neither declares nor installs them, and every skill works without them:

Extension	Adds	Install
pi-subagents	A subagent tool for delegation and parallel child agents	pi install npm:pi-subagents
@juicesharp/rpiv-ask-user-question	A structured questionnaire tool (typed options instead of free-form replies)	pi install npm:@juicesharp/rpiv-ask-user-question

For a first-pass PR review on any platform, use the code-review skill rather than a platform's built-in review command.

just check-skills (also the skill-portability prek hook) enforces this: it fails when a skill names a Claude-only capability (the Agent or Task tool, AskUserQuestion, ToolSearch, TodoWrite, a built-in /review, subagents) in a passage that doesn't also give the fallback. A deliberately platform-specific passage can opt out with <!-- portability: ok -->. See bin/check_skill_portability.py for the exact rule.

Authoring a Skill

1. Pick or create a category directory: skills/<category>/

2. Create skills/<category>/<skill-name>/SKILL.md with frontmatter and body:

   ---
   name: your-skill-name        # must match directory name exactly
   description: >               # what it does AND when to use it (max 1024 chars)
     Use this skill when...
   license: BSD-3-Clause
   metadata:
     category: <category>
   ---
   
   # Skill Title
   
   ...

3. Add the skill to your category's README.md table and to the All Skills table above.

4. Register it in agent-config.toml's [skills] table (and the relevant [profiles.*] entry) so agent-kit apply picks it up.

5. Open a PR — see the repo-level CONTRIBUTING guide.