Skip to main content

Utility Skills

Specialist skills for the tasks that none of the 20 members owns.

The Skills Reference documents the Society members: the 20 roles the runtime can invoke with npx agenthood run. This page documents the utility skills: specialist capability files that any agent runtime loads from skills/, ported from the GitHub Copilot customization library and the deepagents project. They have no runtime invocation (they are not registered members); they activate when their description matches the task at hand.


Teaching & Explaining

Concept Explainer

Most concepts have a simple core under the jargon.

What it does: Breaks down complex technical concepts in beginner-friendly ways: everyday analogies first, terminology second, working examples always, with comprehension checks before moving on.

When to use: Explaining a technical topic to a learner, onboarding someone into unfamiliar code, writing educational documentation, answering "why does this work" with more than a one-liner.

File: skills/concept-explainer/SKILL.md
Relation: Complements The Librarian (documentation): the Explainer teaches, the Librarian records.

Onboarding Planner

Foundation, exploration, integration: a path from first day to first contribution.

What it does: Creates personalized phased onboarding plans: environment setup with troubleshooting, codebase discovery with beginner-friendly first tasks, and team-process integration through early wins.

When to use: A new team member joins, a developer arrives with an unfamiliar stack, or onboarding documentation needs structure.

File: skills/onboarding-plan/SKILL.md
Relation: Complements The Librarian (docs) and The Concept Explainer (learning material).


Planning & Building

Cleanup Specialist

Simplifies safely. Nothing observable changes.

What it does: Removes dead code, consolidates duplication, simplifies messy patterns, and applies consistent formatting across code and documentation: one improvement at a time, tested before and after, never adding features.

When to use: A file, directory, or codebase has accumulated dead code or duplication and needs maintainability work without behavior change.

File: skills/cleanup-specialist/SKILL.md
Relation: Complements The Warden (detects smells): the Warden reports, the Specialist remediates.


GitHub Workflow

Issue Manager

An issue that requires a follow-up conversation to understand has failed.

What it does: Turns vague complaints into workable tickets: bug report essentials (reproduction steps, expected vs actual, environment), structured feature requests (problem, use cases, success criteria), and triage best practices (labels, linking, clarifying questions).

When to use: Filing bug reports or feature requests, triaging an issue queue, or teaching contributors to write actionable issues.

File: skills/issue-manager/SKILL.md
Relation: Complements The Doorman (standards enforcement) and The Scribe (communication): no lane overlap.


Quality

Accessibility Auditor

A page that only works with a mouse is a page that does not work.

What it does: Audits web interfaces against WCAG: semantic HTML first, ARIA requirements, keyboard navigation with visible focus, contrast ratios (4.5:1 text / 3:1 large text and UI), screen reader compatibility, accessible forms, and automated plus manual testing integration.

When to use: Generating or reviewing HTML, auditing an interface against accessibility standards, or designing forms and dynamic content.

File: skills/accessibility-auditor/SKILL.md
Relation: Complements The Auditor (security): a specialized, WCAG-focused review lane.


Memory

Rememberer

Preferences in memory, processes in skills. Unclassified knowledge is unfindable knowledge.

What it does: Scans conversation history, classifies learnings (best practices, anti-patterns, conventions, decision rationale) into memory entries for guidelines or skill files for reusable workflows, and summarizes what was captured and where.

When to use: "Remember this", "save what we learned", "update memory", or "capture learnings": any session that produced knowledge worth keeping.

File: skills/remember/SKILL.md
Relation: Complements The Librarian (knowledge management): the Rememberer captures from sessions, the Librarian curates the corpus.


Infrastructure & Integrations

AWS

The cloud is one command away. Knowing which command is the work.

What it does: Wraps the aws CLI for S3 listing, copy and sync, EC2 instance listing, start and stop, and Lambda listing and invocation. Reads the default profile from ~/.aws/config unless AWS_PROFILE is set.

When to use: Managing S3, EC2, Lambda or other AWS services from the terminal.

File: skills/aws/SKILL.md

Docker

A container that runs on your machine is not yet a container that runs anywhere.

What it does: Images (list, pull, build, remove), containers (run, stop, list), Compose, and cleanup commands. Notes that the daemon must be running and that --rm removes a container on exit.

When to use: Building, running, or debugging containers.

File: skills/docker/SKILL.md
Relation: The Operator deploys and watches the running system; this skill supplies the container commands it works through.

Kubernetes

The pod is the smallest thing you can inspect. The deployment is the thing that matters.

What it does: kubectl commands for pods (list, describe, logs, exec), deployments (list, scale, rollout status, rollback), services, and contexts. Namespace targeting uses -n.

When to use: Deploying, inspecting, or debugging Kubernetes resources.

File: skills/kubernetes/SKILL.md

Datadog

A metric nobody queries is a metric nobody has.

What it does: Datadog HTTP API calls for metrics, monitors, and logs, authenticated with DD_API_KEY. The EU site uses api.datadoghq.eu instead of api.datadoghq.com.

When to use: Querying metrics or logs, or managing monitors.

File: skills/datadog/SKILL.md

Sentry

The stack trace is the beginning of the question, not the answer.

What it does: Uses the hosted Sentry MCP for triage (find organizations and projects, search issues sorted by impact, fetch issue details) and sentry-cli for releases and source maps. Escalation to Seer follows the criteria in docs/portals/sentry.md.

When to use: Triaging production errors or managing releases.

File: skills/sentry/SKILL.md
Relation: The Debugger maps its five-step protocol onto these tools; setup lives in docs/portals/sentry.md.

Elasticsearch

Full text search is a query problem before it is a cluster problem.

What it does: Elasticsearch REST calls for cluster state, search, and index management on the default port 9200. ?pretty formats JSON and /_cat/ endpoints give human-readable output.

When to use: Querying, indexing, or managing Elasticsearch indices.

File: skills/elasticsearch/SKILL.md

MongoDB

Inspect before you query. The schema is whatever the documents say it is.

What it does: mongosh commands for connection, database and collection inspection, and queries. Notes that mongosh replaces the deprecated mongo shell, and gives the connection string format.

When to use: Querying, inspecting, or managing MongoDB collections.

File: skills/mongodb/SKILL.md

MySQL

Credentials belong in a config file, not in a shell history.

What it does: mysql commands for connection, schema inspection, and operations, with credentials stored in ~/.my.cnf rather than passed on the command line.

When to use: Querying, inspecting schema, or managing MySQL databases.

File: skills/mysql/SKILL.md

PostgreSQL

\d answers in one line what a schema dump answers in a hundred.

What it does: psql commands for connection through DATABASE_URL, inspection (\l, \dt, \d, \dn), and one-shot queries with -c.

When to use: Querying, inspecting schema, or managing PostgreSQL databases.

File: skills/postgres/SKILL.md

Redis

KEYS * is a production incident waiting for a large keyspace.

What it does: redis-cli commands for connection, authentication, key operations (get, set, delete, exists, set with expiry), and inspection. Warns that KEYS * blocks and SCAN is the alternative for large datasets.

When to use: Interacting with Redis caches, queues, or key-value stores.

File: skills/redis/SKILL.md

GitHub

The API is the interface. The browser is a convenience.

What it does: gh commands for issues, pull requests (list, view, create, checkout), repositories, and releases. Authentication comes from gh auth login, and --json gives machine-readable output.

When to use: Working with issues, PRs, releases, or repository settings.

File: skills/github/SKILL.md
Relation: The Scribe writes the PR descriptions and The Doorman validates PR standards; the API calls are this skill's.

GitLab

Merge requests, pipelines and issues share one CLI.

What it does: glab commands for merge requests (list, view, create, merge, approve), issues, CI/CD pipelines, and repositories. Authentication comes from glab auth login.

When to use: Working with merge requests, issues, or CI/CD pipelines.

File: skills/gitlab/SKILL.md

Jira

A sprint you cannot query is a sprint you cannot plan.

What it does: jira-cli commands for issues (view, create, update), sprints, and search. The API token comes from Atlassian, and config lives in ~/.jira/.config.yml.

When to use: Viewing, creating, or updating Jira issues.

File: skills/jira/SKILL.md

Linear

The issue tracker is only useful when the terminal can reach it.

What it does: Linear CLI commands for issues, teams and projects, and search, with authentication through linear auth.

When to use: Viewing, creating, or updating Linear tasks.

File: skills/linear/SKILL.md

Email

A notification nobody receives is a notification that did not happen.

What it does: Sends mail through sendmail or msmtp, or over SMTP with curl and a message file that carries the headers. Notes that msmtp reads ~/.msmtprc and that SMTP credentials never belong in a script.

When to use: Sending transactional emails or notifications programmatically.

File: skills/email/SKILL.md

Telegram

A bot token and a chat id are the whole setup.

What it does: Telegram Bot API calls for sending messages, with optional Markdown formatting, plus getUpdates and getMe. Uses TELEGRAM_BOT_TOKEN and TELEGRAM_CHAT_ID.

When to use: Sending notifications or building chat interactions.

File: skills/telegram/SKILL.md

Web Research

An answer without a trail is a rumour with better formatting.

What it does: Turns a question into a plan of 3 to 5 subtopics, gives each subtopic its own parallel search agent with an expected source type, saves findings with URLs and dates under research/, and synthesises one cited report with the answer, numbered findings, disagreements, and gaps. It records a claim as unverified rather than dropping it.

When to use: A question depends on sources outside the repository, needs several independent angles, or will be acted on by someone who must check the evidence afterwards.

File: skills/web-research/SKILL.md



One Society. Each member knows its lane, and each skill knows its task.