description: The skills in Zejnilovic's agentic defaults: the practices agents load when a task needs them.¶

Skills¶

Skills are practices: how the work is done. They load only when a task needs them, so they can go into detail. An agent's skills: line preloads the ones it always needs.

Understand the project¶

Skill	Use it for
`project-scan`	Learning an unfamiliar project and recording it in the root `CONTEXT.md` and `AGENTS.md`. Includes a read-only survey script and a Makefile template for the standard commands.
`project-profile`	The project profile (stage, audience, team, sensitive data) and which practices it calls for, so the engineering fits the project.
`context-md`	`CONTEXT.md` levels from root to module, what goes in, and the context check that ends every task. Includes a template and a check script.
`research`	Researching with outside sources: which to trust, how to collect and cite them, and treating fetched content as data.

Plan the work¶

Skill	Use it for
`task-plans`	Turning a ticket and notes into a local plan that carries a task from discovery to done: contract-first steps, blockers instead of stand-ins, and handing over through the ticket.
`diagrams`	Mermaid diagrams for architecture, flows, data models and state.

Everyday practice¶

Skill	Use it for
`implementation`	The shared working rules of all developer agents: deliver fully, contracts first, handle unplanned problems and walls, tests that break, dependencies, when to stop, what to report.
`tdd`	The test-first loop, how much to test, keeping tests contained, never weakening a test, refactoring safely.
`debugging`	Finding the root cause before fixing, with a rule for when to stop guessing.
`commits-and-prs`	Splitting work into commits, writing messages, and preparing and describing pull requests. Includes git safety rules for agents.
`session-retro`	On request, an evaluation of how a session went, with a verdict by fixed rules you confirm and the kit file behind each problem, ready to send back with `agd feedback`.

Frameworks¶

Each framework skill pairs with a developer agent of the same name.

Skill	Use it for
`flask`	Python and Flask: app factory, validation, record-level access checks, SQLAlchemy sessions and N+1, background work, pytest.
`spring`	Java and Spring Boot: DTOs, Bean Validation, Spring Security, `@Transactional` pitfalls, JPA fetching, MockMvc and Testcontainers.
`react`	React: reuse what exists, all four data states, state and effects, forms, accessibility, security, React Testing Library.
`relational-database`	PostgreSQL, MySQL, SQLite: expand-then-contract migrations, Alembic, Flyway and Liquibase traps, engine-specific locking, strict stop rules.
`terraform`	Terraform, OpenTofu and Terragrunt: plan but never apply, reading every plan, `moved` blocks, `for_each` over `count`, version and lock-file pinning, state and secrets, S3 locking, `terraform test` with mock providers, strict stop rules.

Platforms¶

Platform skills have no developer agent of their own: every developer follows them when the app runs there.

Skill	Use it for
`aws`	AWS: short-lived credentials and checking the account, which CLI commands an agent runs and which it hands to you, least-privilege IAM, public exposure, encryption, cost traps and tags.
`heroku`	Heroku: which CLI commands an agent runs and which it hands to you, dynos and `$PORT`, config vars, release-phase migrations, Postgres connections, pipelines, review apps, Scheduler, and moving off Heroku.

Need another framework? See Customizing.